Spaces:
Runtime error
Runtime error
Upload 60 files
Browse filesThis view is limited to 50 files because it contains too many changes. Β
See raw diff
- models/2.py +22 -0
- models/A2Attention.py +56 -0
- models/CBAM.py +76 -0
- models/EMA.py +31 -0
- models/LSKblock.py +29 -0
- models/S2Attention.py +67 -0
- models/SE.py +39 -0
- models/SGE.py +54 -0
- models/SK.py +53 -0
- models/ShuffleAttention.py +71 -0
- models/SimAM.py +29 -0
- models/__init__.py +0 -0
- models/__pycache__/A2Attention.cpython-38.pyc +0 -0
- models/__pycache__/CBAM.cpython-38.pyc +0 -0
- models/__pycache__/EMA.cpython-38.pyc +0 -0
- models/__pycache__/LSKblock.cpython-38.pyc +0 -0
- models/__pycache__/S2Attention.cpython-38.pyc +0 -0
- models/__pycache__/SE.cpython-38.pyc +0 -0
- models/__pycache__/SGE.cpython-38.pyc +0 -0
- models/__pycache__/SK.cpython-38.pyc +0 -0
- models/__pycache__/ShuffleAttention.cpython-38.pyc +0 -0
- models/__pycache__/SimAM.cpython-38.pyc +0 -0
- models/__pycache__/__init__.cpython-38.pyc +0 -0
- models/__pycache__/common.cpython-38.pyc +0 -0
- models/__pycache__/experimental.cpython-38.pyc +0 -0
- models/__pycache__/yolo.cpython-38.pyc +0 -0
- models/common.py +918 -0
- models/experimental.py +111 -0
- models/hub/anchors.yaml +59 -0
- models/hub/yolov3-spp.yaml +51 -0
- models/hub/yolov3-tiny.yaml +41 -0
- models/hub/yolov3.yaml +51 -0
- models/hub/yolov5-bifpn.yaml +48 -0
- models/hub/yolov5-fpn.yaml +42 -0
- models/hub/yolov5-p2.yaml +54 -0
- models/hub/yolov5-p34.yaml +41 -0
- models/hub/yolov5-p6.yaml +56 -0
- models/hub/yolov5-p7.yaml +67 -0
- models/hub/yolov5-panet.yaml +48 -0
- models/hub/yolov5l6.yaml +60 -0
- models/hub/yolov5m6.yaml +60 -0
- models/hub/yolov5n6.yaml +60 -0
- models/hub/yolov5s-LeakyReLU.yaml +49 -0
- models/hub/yolov5s-ghost.yaml +48 -0
- models/hub/yolov5s-transformer.yaml +48 -0
- models/hub/yolov5s6.yaml +60 -0
- models/hub/yolov5x6.yaml +60 -0
- models/segment/yolov5l-seg.yaml +48 -0
- models/segment/yolov5m-seg.yaml +48 -0
- models/segment/yolov5n-seg.yaml +48 -0
models/2.py
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import gradio as gr
|
3 |
+
|
4 |
+
model = torch.hub.load('', 'custom', path="best.pt", source='local',force_reload=True)
|
5 |
+
|
6 |
+
|
7 |
+
title = " ηηΌηΌΊι·ζ£ζ΅ "
|
8 |
+
|
9 |
+
desc = ""
|
10 |
+
|
11 |
+
base_conf, base_iou = 0.30, 0.45
|
12 |
+
|
13 |
+
def det_image(img, conf_thres, iou_thres):
|
14 |
+
model.conf = conf_thres
|
15 |
+
model.iou = iou_thres
|
16 |
+
return model(img).render()[0]
|
17 |
+
|
18 |
+
gr.Interface(inputs=["image", gr.Slider(minimum=0,maximum=1, value=base_conf), gr.Slider(minimum=0,maximum=1, value=base_iou)],
|
19 |
+
outputs=["image"],
|
20 |
+
fn=det_image,
|
21 |
+
title=title,
|
22 |
+
description=desc).launch()
|
models/A2Attention.py
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import torch
|
3 |
+
from torch import nn
|
4 |
+
from torch.nn import init
|
5 |
+
from torch.nn import functional as F
|
6 |
+
|
7 |
+
|
8 |
+
|
9 |
+
class DoubleAttention(nn.Module):
|
10 |
+
|
11 |
+
def __init__(self, in_channels,c_m=128,c_n=128,reconstruct = True):
|
12 |
+
super().__init__()
|
13 |
+
self.in_channels=in_channels
|
14 |
+
self.reconstruct = reconstruct
|
15 |
+
self.c_m=c_m
|
16 |
+
self.c_n=c_n
|
17 |
+
self.convA=nn.Conv2d(in_channels,c_m,1)
|
18 |
+
self.convB=nn.Conv2d(in_channels,c_n,1)
|
19 |
+
self.convV=nn.Conv2d(in_channels,c_n,1)
|
20 |
+
if self.reconstruct:
|
21 |
+
self.conv_reconstruct = nn.Conv2d(c_m, in_channels, kernel_size = 1)
|
22 |
+
self.init_weights()
|
23 |
+
|
24 |
+
|
25 |
+
def init_weights(self):
|
26 |
+
for m in self.modules():
|
27 |
+
if isinstance(m, nn.Conv2d):
|
28 |
+
init.kaiming_normal_(m.weight, mode='fan_out')
|
29 |
+
if m.bias is not None:
|
30 |
+
init.constant_(m.bias, 0)
|
31 |
+
elif isinstance(m, nn.BatchNorm2d):
|
32 |
+
init.constant_(m.weight, 1)
|
33 |
+
init.constant_(m.bias, 0)
|
34 |
+
elif isinstance(m, nn.Linear):
|
35 |
+
init.normal_(m.weight, std=0.001)
|
36 |
+
if m.bias is not None:
|
37 |
+
init.constant_(m.bias, 0)
|
38 |
+
|
39 |
+
def forward(self, x):
|
40 |
+
b, c, h,w=x.shape
|
41 |
+
assert c==self.in_channels
|
42 |
+
A=self.convA(x) #b,c_m,h,w
|
43 |
+
B=self.convB(x) #b,c_n,h,w
|
44 |
+
V=self.convV(x) #b,c_n,h,w
|
45 |
+
tmpA=A.view(b,self.c_m,-1)
|
46 |
+
attention_maps=F.softmax(B.view(b,self.c_n,-1))
|
47 |
+
attention_vectors=F.softmax(V.view(b,self.c_n,-1))
|
48 |
+
# step 1: feature gating
|
49 |
+
global_descriptors=torch.bmm(tmpA,attention_maps.permute(0,2,1)) #b.c_m,c_n
|
50 |
+
# step 2: feature distribution
|
51 |
+
tmpZ = global_descriptors.matmul(attention_vectors) #b,c_m,h*w
|
52 |
+
tmpZ=tmpZ.view(b,self.c_m,h,w) #b,c_m,h,w
|
53 |
+
if self.reconstruct:
|
54 |
+
tmpZ=self.conv_reconstruct(tmpZ)
|
55 |
+
|
56 |
+
return tmpZ
|
models/CBAM.py
ADDED
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import torch
|
3 |
+
from torch import nn
|
4 |
+
from torch.nn import init
|
5 |
+
|
6 |
+
|
7 |
+
class ChannelAttention(nn.Module):
|
8 |
+
def __init__(self, channel, reduction=16):
|
9 |
+
super().__init__()
|
10 |
+
self.maxpool = nn.AdaptiveMaxPool2d(1)
|
11 |
+
self.avgpool = nn.AdaptiveAvgPool2d(1)
|
12 |
+
self.se = nn.Sequential(
|
13 |
+
nn.Conv2d(channel, channel // reduction, 1, bias=False),
|
14 |
+
nn.ReLU(),
|
15 |
+
nn.Conv2d(channel // reduction, channel, 1, bias=False)
|
16 |
+
)
|
17 |
+
self.sigmoid = nn.Sigmoid()
|
18 |
+
|
19 |
+
def forward(self, x):
|
20 |
+
max_result = self.maxpool(x)
|
21 |
+
avg_result = self.avgpool(x)
|
22 |
+
max_out = self.se(max_result)
|
23 |
+
avg_out = self.se(avg_result)
|
24 |
+
output = self.sigmoid(max_out + avg_out)
|
25 |
+
return output
|
26 |
+
|
27 |
+
|
28 |
+
class SpatialAttention(nn.Module):
|
29 |
+
def __init__(self, kernel_size=7):
|
30 |
+
super().__init__()
|
31 |
+
self.conv = nn.Conv2d(2, 1, kernel_size=kernel_size, padding=kernel_size // 2)
|
32 |
+
self.sigmoid = nn.Sigmoid()
|
33 |
+
|
34 |
+
def forward(self, x):
|
35 |
+
max_result, _ = torch.max(x, dim=1, keepdim=True)
|
36 |
+
avg_result = torch.mean(x, dim=1, keepdim=True)
|
37 |
+
result = torch.cat([max_result, avg_result], 1)
|
38 |
+
output = self.conv(result)
|
39 |
+
output = self.sigmoid(output)
|
40 |
+
return output
|
41 |
+
|
42 |
+
|
43 |
+
class CBAMBlock(nn.Module):
|
44 |
+
|
45 |
+
def __init__(self, channel=512, reduction=16, kernel_size=7):
|
46 |
+
super().__init__()
|
47 |
+
self.ca = ChannelAttention(channel=channel, reduction=reduction)
|
48 |
+
self.sa = SpatialAttention(kernel_size=kernel_size)
|
49 |
+
|
50 |
+
def init_weights(self):
|
51 |
+
for m in self.modules():
|
52 |
+
if isinstance(m, nn.Conv2d):
|
53 |
+
init.kaiming_normal_(m.weight, mode='fan_out')
|
54 |
+
if m.bias is not None:
|
55 |
+
init.constant_(m.bias, 0)
|
56 |
+
elif isinstance(m, nn.BatchNorm2d):
|
57 |
+
init.constant_(m.weight, 1)
|
58 |
+
init.constant_(m.bias, 0)
|
59 |
+
elif isinstance(m, nn.Linear):
|
60 |
+
init.normal_(m.weight, std=0.001)
|
61 |
+
if m.bias is not None:
|
62 |
+
init.constant_(m.bias, 0)
|
63 |
+
|
64 |
+
def forward(self, x):
|
65 |
+
b, c, _, _ = x.size()
|
66 |
+
out = x * self.ca(x)
|
67 |
+
out = out * self.sa(out)
|
68 |
+
return out
|
69 |
+
|
70 |
+
|
71 |
+
if __name__ == '__main__':
|
72 |
+
input = torch.randn(50, 512, 7, 7)
|
73 |
+
kernel_size = input.shape[2]
|
74 |
+
cbam = CBAMBlock(channel=512, reduction=16, kernel_size=kernel_size)
|
75 |
+
output = cbam(input)
|
76 |
+
print(output.shape)
|
models/EMA.py
ADDED
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch import nn
|
3 |
+
|
4 |
+
class EMA(nn.Module):
|
5 |
+
def __init__(self, channels, factor=8):
|
6 |
+
super(EMA, self).__init__()
|
7 |
+
self.groups = factor
|
8 |
+
assert channels // self.groups > 0
|
9 |
+
self.softmax = nn.Softmax(-1)
|
10 |
+
self.agp = nn.AdaptiveAvgPool2d((1, 1))
|
11 |
+
self.pool_h = nn.AdaptiveAvgPool2d((None, 1))
|
12 |
+
self.pool_w = nn.AdaptiveAvgPool2d((1, None))
|
13 |
+
self.gn = nn.GroupNorm(channels // self.groups, channels // self.groups)
|
14 |
+
self.conv1x1 = nn.Conv2d(channels // self.groups, channels // self.groups, kernel_size=1, stride=1, padding=0)
|
15 |
+
self.conv3x3 = nn.Conv2d(channels // self.groups, channels // self.groups, kernel_size=3, stride=1, padding=1)
|
16 |
+
|
17 |
+
def forward(self, x):
|
18 |
+
b, c, h, w = x.size()
|
19 |
+
group_x = x.reshape(b * self.groups, -1, h, w) # b*g,c//g,h,w
|
20 |
+
x_h = self.pool_h(group_x)
|
21 |
+
x_w = self.pool_w(group_x).permute(0, 1, 3, 2)
|
22 |
+
hw = self.conv1x1(torch.cat([x_h, x_w], dim=2))
|
23 |
+
x_h, x_w = torch.split(hw, [h, w], dim=2)
|
24 |
+
x1 = self.gn(group_x * x_h.sigmoid() * x_w.permute(0, 1, 3, 2).sigmoid())
|
25 |
+
x2 = self.conv3x3(group_x)
|
26 |
+
x11 = self.softmax(self.agp(x1).reshape(b * self.groups, -1, 1).permute(0, 2, 1))
|
27 |
+
x12 = x2.reshape(b * self.groups, c // self.groups, -1) # b*g, c//g, hw
|
28 |
+
x21 = self.softmax(self.agp(x2).reshape(b * self.groups, -1, 1).permute(0, 2, 1))
|
29 |
+
x22 = x1.reshape(b * self.groups, c // self.groups, -1) # b*g, c//g, hw
|
30 |
+
weights = (torch.matmul(x11, x12) + torch.matmul(x21, x22)).reshape(b * self.groups, 1, h, w)
|
31 |
+
return (group_x * weights.sigmoid()).reshape(b, c, h, w)
|
models/LSKblock.py
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn as nn
|
3 |
+
|
4 |
+
|
5 |
+
class LSKblock(nn.Module):
|
6 |
+
def __init__(self, dim):
|
7 |
+
super().__init__()
|
8 |
+
self.conv0 = nn.Conv2d(dim, dim, 5, padding=2, groups=dim)
|
9 |
+
self.conv_spatial = nn.Conv2d(dim, dim, 7, stride=1, padding=9, groups=dim, dilation=3)
|
10 |
+
self.conv1 = nn.Conv2d(dim, dim // 2, 1)
|
11 |
+
self.conv2 = nn.Conv2d(dim, dim // 2, 1)
|
12 |
+
self.conv_squeeze = nn.Conv2d(2, 2, 7, padding=3)
|
13 |
+
self.conv = nn.Conv2d(dim // 2, dim, 1)
|
14 |
+
|
15 |
+
def forward(self, x):
|
16 |
+
attn1 = self.conv0(x)
|
17 |
+
attn2 = self.conv_spatial(attn1)
|
18 |
+
|
19 |
+
attn1 = self.conv1(attn1)
|
20 |
+
attn2 = self.conv2(attn2)
|
21 |
+
|
22 |
+
attn = torch.cat([attn1, attn2], dim=1)
|
23 |
+
avg_attn = torch.mean(attn, dim=1, keepdim=True)
|
24 |
+
max_attn, _ = torch.max(attn, dim=1, keepdim=True)
|
25 |
+
agg = torch.cat([avg_attn, max_attn], dim=1)
|
26 |
+
sig = self.conv_squeeze(agg).sigmoid()
|
27 |
+
attn = attn1 * sig[:, 0, :, :].unsqueeze(1) + attn2 * sig[:, 1, :, :].unsqueeze(1)
|
28 |
+
attn = self.conv(attn)
|
29 |
+
return x * attn
|
models/S2Attention.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import torch
|
3 |
+
from torch import nn
|
4 |
+
from torch.nn import init
|
5 |
+
|
6 |
+
|
7 |
+
def spatial_shift1(x):
|
8 |
+
b, w, h, c = x.size()
|
9 |
+
x[:, 1:, :, :c // 4] = x[:, :w - 1, :, :c // 4]
|
10 |
+
x[:, :w - 1, :, c // 4:c // 2] = x[:, 1:, :, c // 4:c // 2]
|
11 |
+
x[:, :, 1:, c // 2:c * 3 // 4] = x[:, :, :h - 1, c // 2:c * 3 // 4]
|
12 |
+
x[:, :, :h - 1, 3 * c // 4:] = x[:, :, 1:, 3 * c // 4:]
|
13 |
+
return x
|
14 |
+
|
15 |
+
|
16 |
+
def spatial_shift2(x):
|
17 |
+
b, w, h, c = x.size()
|
18 |
+
x[:, :, 1:, :c // 4] = x[:, :, :h - 1, :c // 4]
|
19 |
+
x[:, :, :h - 1, c // 4:c // 2] = x[:, :, 1:, c // 4:c // 2]
|
20 |
+
x[:, 1:, :, c // 2:c * 3 // 4] = x[:, :w - 1, :, c // 2:c * 3 // 4]
|
21 |
+
x[:, :w - 1, :, 3 * c // 4:] = x[:, 1:, :, 3 * c // 4:]
|
22 |
+
return x
|
23 |
+
|
24 |
+
|
25 |
+
class SplitAttention(nn.Module):
|
26 |
+
def __init__(self, channel=512, k=3):
|
27 |
+
super().__init__()
|
28 |
+
self.channel = channel
|
29 |
+
self.k = k
|
30 |
+
self.mlp1 = nn.Linear(channel, channel, bias=False)
|
31 |
+
self.gelu = nn.GELU()
|
32 |
+
self.mlp2 = nn.Linear(channel, channel * k, bias=False)
|
33 |
+
self.softmax = nn.Softmax(1)
|
34 |
+
|
35 |
+
def forward(self, x_all):
|
36 |
+
b, k, h, w, c = x_all.shape
|
37 |
+
x_all = x_all.reshape(b, k, -1, c) # bs,k,n,c
|
38 |
+
a = torch.sum(torch.sum(x_all, 1), 1) # bs,c
|
39 |
+
hat_a = self.mlp2(self.gelu(self.mlp1(a))) # bs,kc
|
40 |
+
hat_a = hat_a.reshape(b, self.k, c) # bs,k,c
|
41 |
+
bar_a = self.softmax(hat_a) # bs,k,c
|
42 |
+
attention = bar_a.unsqueeze(-2) # #bs,k,1,c
|
43 |
+
out = attention * x_all # #bs,k,n,c
|
44 |
+
out = torch.sum(out, 1).reshape(b, h, w, c)
|
45 |
+
return out
|
46 |
+
|
47 |
+
|
48 |
+
class S2Attention(nn.Module):
|
49 |
+
|
50 |
+
def __init__(self, channels=512):
|
51 |
+
super().__init__()
|
52 |
+
self.mlp1 = nn.Linear(channels, channels * 3)
|
53 |
+
self.mlp2 = nn.Linear(channels, channels)
|
54 |
+
self.split_attention = SplitAttention()
|
55 |
+
|
56 |
+
def forward(self, x):
|
57 |
+
b, c, w, h = x.size()
|
58 |
+
x = x.permute(0, 2, 3, 1)
|
59 |
+
x = self.mlp1(x)
|
60 |
+
x1 = spatial_shift1(x[:, :, :, :c])
|
61 |
+
x2 = spatial_shift2(x[:, :, :, c:c * 2])
|
62 |
+
x3 = x[:, :, :, c * 2:]
|
63 |
+
x_all = torch.stack([x1, x2, x3], 1)
|
64 |
+
a = self.split_attention(x_all)
|
65 |
+
x = self.mlp2(a)
|
66 |
+
x = x.permute(0, 3, 1, 2)
|
67 |
+
return x
|
models/SE.py
ADDED
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import torch
|
3 |
+
from torch import nn
|
4 |
+
from torch.nn import init
|
5 |
+
|
6 |
+
|
7 |
+
|
8 |
+
class SEAttention(nn.Module):
|
9 |
+
|
10 |
+
def __init__(self, channel=512,reduction=16):
|
11 |
+
super().__init__()
|
12 |
+
self.avg_pool = nn.AdaptiveAvgPool2d(1)
|
13 |
+
self.fc = nn.Sequential(
|
14 |
+
nn.Linear(channel, channel // reduction, bias=False),
|
15 |
+
nn.ReLU(inplace=True),
|
16 |
+
nn.Linear(channel // reduction, channel, bias=False),
|
17 |
+
nn.Sigmoid()
|
18 |
+
)
|
19 |
+
|
20 |
+
|
21 |
+
def init_weights(self):
|
22 |
+
for m in self.modules():
|
23 |
+
if isinstance(m, nn.Conv2d):
|
24 |
+
init.kaiming_normal_(m.weight, mode='fan_out')
|
25 |
+
if m.bias is not None:
|
26 |
+
init.constant_(m.bias, 0)
|
27 |
+
elif isinstance(m, nn.BatchNorm2d):
|
28 |
+
init.constant_(m.weight, 1)
|
29 |
+
init.constant_(m.bias, 0)
|
30 |
+
elif isinstance(m, nn.Linear):
|
31 |
+
init.normal_(m.weight, std=0.001)
|
32 |
+
if m.bias is not None:
|
33 |
+
init.constant_(m.bias, 0)
|
34 |
+
|
35 |
+
def forward(self, x):
|
36 |
+
b, c, _, _ = x.size()
|
37 |
+
y = self.avg_pool(x).view(b, c)
|
38 |
+
y = self.fc(y).view(b, c, 1, 1)
|
39 |
+
return x * y.expand_as(x)
|
models/SGE.py
ADDED
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import torch
|
3 |
+
from torch import nn
|
4 |
+
from torch.nn import init
|
5 |
+
|
6 |
+
|
7 |
+
class SpatialGroupEnhance(nn.Module):
|
8 |
+
def __init__(self, groups=8):
|
9 |
+
super().__init__()
|
10 |
+
self.groups = groups
|
11 |
+
self.avg_pool = nn.AdaptiveAvgPool2d(1)
|
12 |
+
self.weight = nn.Parameter(torch.zeros(1, groups, 1, 1))
|
13 |
+
self.bias = nn.Parameter(torch.zeros(1, groups, 1, 1))
|
14 |
+
self.sig = nn.Sigmoid()
|
15 |
+
self.init_weights()
|
16 |
+
|
17 |
+
def init_weights(self):
|
18 |
+
for m in self.modules():
|
19 |
+
if isinstance(m, nn.Conv2d):
|
20 |
+
init.kaiming_normal_(m.weight, mode='fan_out')
|
21 |
+
if m.bias is not None:
|
22 |
+
init.constant_(m.bias, 0)
|
23 |
+
elif isinstance(m, nn.BatchNorm2d):
|
24 |
+
init.constant_(m.weight, 1)
|
25 |
+
init.constant_(m.bias, 0)
|
26 |
+
elif isinstance(m, nn.Linear):
|
27 |
+
init.normal_(m.weight, std=0.001)
|
28 |
+
if m.bias is not None:
|
29 |
+
init.constant_(m.bias, 0)
|
30 |
+
|
31 |
+
def forward(self, x):
|
32 |
+
b, c, h, w = x.shape
|
33 |
+
x = x.view(b * self.groups, -1, h, w) # bs*g,dim//g,h,w
|
34 |
+
xn = x * self.avg_pool(x) # bs*g,dim//g,h,w
|
35 |
+
xn = xn.sum(dim=1, keepdim=True) # bs*g,1,h,w
|
36 |
+
t = xn.view(b * self.groups, -1) # bs*g,h*w
|
37 |
+
|
38 |
+
t = t - t.mean(dim=1, keepdim=True) # bs*g,h*w
|
39 |
+
std = t.std(dim=1, keepdim=True) + 1e-5
|
40 |
+
t = t / std # bs*g,h*w
|
41 |
+
t = t.view(b, self.groups, h, w) # bs,g,h*w
|
42 |
+
|
43 |
+
t = t * self.weight + self.bias # bs,g,h*w
|
44 |
+
t = t.view(b * self.groups, 1, h, w) # bs*g,1,h*w
|
45 |
+
x = x * self.sig(t)
|
46 |
+
x = x.view(b, c, h, w)
|
47 |
+
return x
|
48 |
+
|
49 |
+
|
50 |
+
if __name__ == '__main__':
|
51 |
+
input = torch.randn(50, 512, 7, 7)
|
52 |
+
sge = SpatialGroupEnhance(groups=8)
|
53 |
+
output = sge(input)
|
54 |
+
print(output.shape)
|
models/SK.py
ADDED
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import torch
|
3 |
+
from torch import nn
|
4 |
+
from torch.nn import init
|
5 |
+
from collections import OrderedDict
|
6 |
+
|
7 |
+
|
8 |
+
class SKAttention(nn.Module):
|
9 |
+
|
10 |
+
def __init__(self, channel=512, kernels=[1,3,5,7], reduction=16, group=1, L=32):
|
11 |
+
super().__init__()
|
12 |
+
self.d = max(L, channel // reduction)
|
13 |
+
self.convs = nn.ModuleList([])
|
14 |
+
for k in kernels:
|
15 |
+
self.convs.append(
|
16 |
+
nn.Sequential(OrderedDict([
|
17 |
+
('conv', nn.Conv2d(channel, channel, kernel_size=k, padding=k // 2, groups=group)),
|
18 |
+
('bn', nn.BatchNorm2d(channel)),
|
19 |
+
('relu', nn.ReLU())
|
20 |
+
]))
|
21 |
+
)
|
22 |
+
self.fc = nn.Linear(channel, self.d)
|
23 |
+
self.fcs = nn.ModuleList([])
|
24 |
+
for i in range(len(kernels)):
|
25 |
+
self.fcs.append(nn.Linear(self.d, channel))
|
26 |
+
self.softmax = nn.Softmax(dim=0)
|
27 |
+
|
28 |
+
def forward(self, x):
|
29 |
+
bs, c, _, _ = x.size()
|
30 |
+
conv_outs = []
|
31 |
+
### split
|
32 |
+
for conv in self.convs:
|
33 |
+
conv_outs.append(conv(x))
|
34 |
+
feats = torch.stack(conv_outs, 0) # k,bs,channel,h,w
|
35 |
+
|
36 |
+
### fuse
|
37 |
+
U = sum(conv_outs) # bs,c,h,w
|
38 |
+
|
39 |
+
### reduction channel
|
40 |
+
S = U.mean(-1).mean(-1) # bs,c
|
41 |
+
Z = self.fc(S) # bs,d
|
42 |
+
|
43 |
+
### calculate attention weight
|
44 |
+
weights = []
|
45 |
+
for fc in self.fcs:
|
46 |
+
weight = fc(Z)
|
47 |
+
weights.append(weight.view(bs, c, 1, 1)) # bs,channel
|
48 |
+
attention_weughts = torch.stack(weights, 0) # k,bs,channel,1,1
|
49 |
+
attention_weughts = self.softmax(attention_weughts) # k,bs,channel,1,1
|
50 |
+
|
51 |
+
### fuse
|
52 |
+
V = (attention_weughts * feats).sum(0)
|
53 |
+
return V
|
models/ShuffleAttention.py
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import torch
|
3 |
+
from torch import nn
|
4 |
+
from torch.nn import init
|
5 |
+
from torch.nn.parameter import Parameter
|
6 |
+
|
7 |
+
|
8 |
+
class ShuffleAttention(nn.Module):
|
9 |
+
|
10 |
+
def __init__(self, channel=512, reduction=16, G=8):
|
11 |
+
super().__init__()
|
12 |
+
self.G = G
|
13 |
+
self.channel = channel
|
14 |
+
self.avg_pool = nn.AdaptiveAvgPool2d(1)
|
15 |
+
self.gn = nn.GroupNorm(channel // (2 * G), channel // (2 * G))
|
16 |
+
self.cweight = Parameter(torch.zeros(1, channel // (2 * G), 1, 1))
|
17 |
+
self.cbias = Parameter(torch.ones(1, channel // (2 * G), 1, 1))
|
18 |
+
self.sweight = Parameter(torch.zeros(1, channel // (2 * G), 1, 1))
|
19 |
+
self.sbias = Parameter(torch.ones(1, channel // (2 * G), 1, 1))
|
20 |
+
self.sigmoid = nn.Sigmoid()
|
21 |
+
|
22 |
+
def init_weights(self):
|
23 |
+
for m in self.modules():
|
24 |
+
if isinstance(m, nn.Conv2d):
|
25 |
+
init.kaiming_normal_(m.weight, mode='fan_out')
|
26 |
+
if m.bias is not None:
|
27 |
+
init.constant_(m.bias, 0)
|
28 |
+
elif isinstance(m, nn.BatchNorm2d):
|
29 |
+
init.constant_(m.weight, 1)
|
30 |
+
init.constant_(m.bias, 0)
|
31 |
+
elif isinstance(m, nn.Linear):
|
32 |
+
init.normal_(m.weight, std=0.001)
|
33 |
+
if m.bias is not None:
|
34 |
+
init.constant_(m.bias, 0)
|
35 |
+
|
36 |
+
@staticmethod
|
37 |
+
def channel_shuffle(x, groups):
|
38 |
+
b, c, h, w = x.shape
|
39 |
+
x = x.reshape(b, groups, -1, h, w)
|
40 |
+
x = x.permute(0, 2, 1, 3, 4)
|
41 |
+
|
42 |
+
# flatten
|
43 |
+
x = x.reshape(b, -1, h, w)
|
44 |
+
|
45 |
+
return x
|
46 |
+
|
47 |
+
def forward(self, x):
|
48 |
+
b, c, h, w = x.size()
|
49 |
+
# group into subfeatures
|
50 |
+
x = x.view(b * self.G, -1, h, w) # bs*G,c//G,h,w
|
51 |
+
|
52 |
+
# channel_split
|
53 |
+
x_0, x_1 = x.chunk(2, dim=1) # bs*G,c//(2*G),h,w
|
54 |
+
|
55 |
+
# channel attention
|
56 |
+
x_channel = self.avg_pool(x_0) # bs*G,c//(2*G),1,1
|
57 |
+
x_channel = self.cweight * x_channel + self.cbias # bs*G,c//(2*G),1,1
|
58 |
+
x_channel = x_0 * self.sigmoid(x_channel)
|
59 |
+
|
60 |
+
# spatial attention
|
61 |
+
x_spatial = self.gn(x_1) # bs*G,c//(2*G),h,w
|
62 |
+
x_spatial = self.sweight * x_spatial + self.sbias # bs*G,c//(2*G),h,w
|
63 |
+
x_spatial = x_1 * self.sigmoid(x_spatial) # bs*G,c//(2*G),h,w
|
64 |
+
|
65 |
+
# concatenate along channel axis
|
66 |
+
out = torch.cat([x_channel, x_spatial], dim=1) # bs*G,c//G,h,w
|
67 |
+
out = out.contiguous().view(b, -1, h, w)
|
68 |
+
|
69 |
+
# channel shuffle
|
70 |
+
out = self.channel_shuffle(out, 2)
|
71 |
+
return out
|
models/SimAM.py
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn as nn
|
3 |
+
|
4 |
+
|
5 |
+
class SimAM(torch.nn.Module):
|
6 |
+
def __init__(self, e_lambda=1e-4):
|
7 |
+
super(SimAM, self).__init__()
|
8 |
+
|
9 |
+
self.activaton = nn.Sigmoid()
|
10 |
+
self.e_lambda = e_lambda
|
11 |
+
|
12 |
+
def __repr__(self):
|
13 |
+
s = self.__class__.__name__ + '('
|
14 |
+
s += ('lambda=%f)' % self.e_lambda)
|
15 |
+
return s
|
16 |
+
|
17 |
+
@staticmethod
|
18 |
+
def get_module_name():
|
19 |
+
return "simam"
|
20 |
+
|
21 |
+
def forward(self, x):
|
22 |
+
b, c, h, w = x.size()
|
23 |
+
|
24 |
+
n = w * h - 1
|
25 |
+
|
26 |
+
x_minus_mu_square = (x - x.mean(dim=[2, 3], keepdim=True)).pow(2)
|
27 |
+
y = x_minus_mu_square / (4 * (x_minus_mu_square.sum(dim=[2, 3], keepdim=True) / n + self.e_lambda)) + 0.5
|
28 |
+
|
29 |
+
return x * self.activaton(y)
|
models/__init__.py
ADDED
File without changes
|
models/__pycache__/A2Attention.cpython-38.pyc
ADDED
Binary file (1.93 kB). View file
|
|
models/__pycache__/CBAM.cpython-38.pyc
ADDED
Binary file (3.02 kB). View file
|
|
models/__pycache__/EMA.cpython-38.pyc
ADDED
Binary file (1.64 kB). View file
|
|
models/__pycache__/LSKblock.cpython-38.pyc
ADDED
Binary file (1.34 kB). View file
|
|
models/__pycache__/S2Attention.cpython-38.pyc
ADDED
Binary file (3 kB). View file
|
|
models/__pycache__/SE.cpython-38.pyc
ADDED
Binary file (1.52 kB). View file
|
|
models/__pycache__/SGE.cpython-38.pyc
ADDED
Binary file (1.91 kB). View file
|
|
models/__pycache__/SK.cpython-38.pyc
ADDED
Binary file (1.63 kB). View file
|
|
models/__pycache__/ShuffleAttention.cpython-38.pyc
ADDED
Binary file (2.34 kB). View file
|
|
models/__pycache__/SimAM.cpython-38.pyc
ADDED
Binary file (1.28 kB). View file
|
|
models/__pycache__/__init__.cpython-38.pyc
ADDED
Binary file (146 Bytes). View file
|
|
models/__pycache__/common.cpython-38.pyc
ADDED
Binary file (38.8 kB). View file
|
|
models/__pycache__/experimental.cpython-38.pyc
ADDED
Binary file (4.86 kB). View file
|
|
models/__pycache__/yolo.cpython-38.pyc
ADDED
Binary file (16.8 kB). View file
|
|
models/common.py
ADDED
@@ -0,0 +1,918 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
"""
|
3 |
+
Common modules
|
4 |
+
"""
|
5 |
+
|
6 |
+
import ast
|
7 |
+
import contextlib
|
8 |
+
import json
|
9 |
+
import math
|
10 |
+
import platform
|
11 |
+
import warnings
|
12 |
+
import zipfile
|
13 |
+
from collections import OrderedDict, namedtuple
|
14 |
+
from copy import copy
|
15 |
+
from pathlib import Path
|
16 |
+
from urllib.parse import urlparse
|
17 |
+
|
18 |
+
import cv2
|
19 |
+
import numpy as np
|
20 |
+
import pandas as pd
|
21 |
+
import requests
|
22 |
+
import torch
|
23 |
+
import torch.nn as nn
|
24 |
+
from PIL import Image
|
25 |
+
from torch.cuda import amp
|
26 |
+
|
27 |
+
# Import 'ultralytics' package or install if if missing
|
28 |
+
try:
|
29 |
+
import ultralytics
|
30 |
+
|
31 |
+
assert hasattr(ultralytics, '__version__') # verify package is not directory
|
32 |
+
except (ImportError, AssertionError):
|
33 |
+
import os
|
34 |
+
|
35 |
+
os.system('pip install -U ultralytics')
|
36 |
+
import ultralytics
|
37 |
+
|
38 |
+
from ultralytics.utils.plotting import Annotator, colors, save_one_box
|
39 |
+
|
40 |
+
from utils import TryExcept
|
41 |
+
from utils.dataloaders import exif_transpose, letterbox
|
42 |
+
from utils.general import (LOGGER, ROOT, Profile, check_requirements, check_suffix, check_version, colorstr,
|
43 |
+
increment_path, is_jupyter, make_divisible, non_max_suppression, scale_boxes, xywh2xyxy,
|
44 |
+
xyxy2xywh, yaml_load)
|
45 |
+
from utils.torch_utils import copy_attr, smart_inference_mode
|
46 |
+
|
47 |
+
|
48 |
+
def autopad(k, p=None, d=1): # kernel, padding, dilation
|
49 |
+
# Pad to 'same' shape outputs
|
50 |
+
if d > 1:
|
51 |
+
k = d * (k - 1) + 1 if isinstance(k, int) else [d * (x - 1) + 1 for x in k] # actual kernel-size
|
52 |
+
if p is None:
|
53 |
+
p = k // 2 if isinstance(k, int) else [x // 2 for x in k] # auto-pad
|
54 |
+
return p
|
55 |
+
|
56 |
+
|
57 |
+
class Conv(nn.Module):
|
58 |
+
# Standard convolution with args(ch_in, ch_out, kernel, stride, padding, groups, dilation, activation)
|
59 |
+
default_act = nn.SiLU() # default activation
|
60 |
+
|
61 |
+
def __init__(self, c1, c2, k=1, s=1, p=None, g=1, d=1, act=True):
|
62 |
+
super().__init__()
|
63 |
+
self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p, d), groups=g, dilation=d, bias=False)
|
64 |
+
self.bn = nn.BatchNorm2d(c2)
|
65 |
+
self.act = self.default_act if act is True else act if isinstance(act, nn.Module) else nn.Identity()
|
66 |
+
|
67 |
+
def forward(self, x):
|
68 |
+
return self.act(self.bn(self.conv(x)))
|
69 |
+
|
70 |
+
def forward_fuse(self, x):
|
71 |
+
return self.act(self.conv(x))
|
72 |
+
|
73 |
+
|
74 |
+
class DWConv(Conv):
|
75 |
+
# Depth-wise convolution
|
76 |
+
def __init__(self, c1, c2, k=1, s=1, d=1, act=True): # ch_in, ch_out, kernel, stride, dilation, activation
|
77 |
+
super().__init__(c1, c2, k, s, g=math.gcd(c1, c2), d=d, act=act)
|
78 |
+
|
79 |
+
|
80 |
+
class DWConvTranspose2d(nn.ConvTranspose2d):
|
81 |
+
# Depth-wise transpose convolution
|
82 |
+
def __init__(self, c1, c2, k=1, s=1, p1=0, p2=0): # ch_in, ch_out, kernel, stride, padding, padding_out
|
83 |
+
super().__init__(c1, c2, k, s, p1, p2, groups=math.gcd(c1, c2))
|
84 |
+
|
85 |
+
|
86 |
+
class TransformerLayer(nn.Module):
|
87 |
+
# Transformer layer https://arxiv.org/abs/2010.11929 (LayerNorm layers removed for better performance)
|
88 |
+
def __init__(self, c, num_heads):
|
89 |
+
super().__init__()
|
90 |
+
self.q = nn.Linear(c, c, bias=False)
|
91 |
+
self.k = nn.Linear(c, c, bias=False)
|
92 |
+
self.v = nn.Linear(c, c, bias=False)
|
93 |
+
self.ma = nn.MultiheadAttention(embed_dim=c, num_heads=num_heads)
|
94 |
+
self.fc1 = nn.Linear(c, c, bias=False)
|
95 |
+
self.fc2 = nn.Linear(c, c, bias=False)
|
96 |
+
|
97 |
+
def forward(self, x):
|
98 |
+
x = self.ma(self.q(x), self.k(x), self.v(x))[0] + x
|
99 |
+
x = self.fc2(self.fc1(x)) + x
|
100 |
+
return x
|
101 |
+
|
102 |
+
|
103 |
+
class TransformerBlock(nn.Module):
|
104 |
+
# Vision Transformer https://arxiv.org/abs/2010.11929
|
105 |
+
def __init__(self, c1, c2, num_heads, num_layers):
|
106 |
+
super().__init__()
|
107 |
+
self.conv = None
|
108 |
+
if c1 != c2:
|
109 |
+
self.conv = Conv(c1, c2)
|
110 |
+
self.linear = nn.Linear(c2, c2) # learnable position embedding
|
111 |
+
self.tr = nn.Sequential(*(TransformerLayer(c2, num_heads) for _ in range(num_layers)))
|
112 |
+
self.c2 = c2
|
113 |
+
|
114 |
+
def forward(self, x):
|
115 |
+
if self.conv is not None:
|
116 |
+
x = self.conv(x)
|
117 |
+
b, _, w, h = x.shape
|
118 |
+
p = x.flatten(2).permute(2, 0, 1)
|
119 |
+
return self.tr(p + self.linear(p)).permute(1, 2, 0).reshape(b, self.c2, w, h)
|
120 |
+
|
121 |
+
|
122 |
+
class Bottleneck(nn.Module):
|
123 |
+
# Standard bottleneck
|
124 |
+
def __init__(self, c1, c2, shortcut=True, g=1, e=0.5): # ch_in, ch_out, shortcut, groups, expansion
|
125 |
+
super().__init__()
|
126 |
+
c_ = int(c2 * e) # hidden channels
|
127 |
+
self.cv1 = Conv(c1, c_, 1, 1)
|
128 |
+
self.cv2 = Conv(c_, c2, 3, 1, g=g)
|
129 |
+
self.add = shortcut and c1 == c2
|
130 |
+
|
131 |
+
def forward(self, x):
|
132 |
+
return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
|
133 |
+
|
134 |
+
|
135 |
+
class BottleneckCSP(nn.Module):
|
136 |
+
# CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks
|
137 |
+
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5): # ch_in, ch_out, number, shortcut, groups, expansion
|
138 |
+
super().__init__()
|
139 |
+
c_ = int(c2 * e) # hidden channels
|
140 |
+
self.cv1 = Conv(c1, c_, 1, 1)
|
141 |
+
self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False)
|
142 |
+
self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False)
|
143 |
+
self.cv4 = Conv(2 * c_, c2, 1, 1)
|
144 |
+
self.bn = nn.BatchNorm2d(2 * c_) # applied to cat(cv2, cv3)
|
145 |
+
self.act = nn.SiLU()
|
146 |
+
self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))
|
147 |
+
|
148 |
+
def forward(self, x):
|
149 |
+
y1 = self.cv3(self.m(self.cv1(x)))
|
150 |
+
y2 = self.cv2(x)
|
151 |
+
return self.cv4(self.act(self.bn(torch.cat((y1, y2), 1))))
|
152 |
+
|
153 |
+
|
154 |
+
class CrossConv(nn.Module):
|
155 |
+
# Cross Convolution Downsample
|
156 |
+
def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False):
|
157 |
+
# ch_in, ch_out, kernel, stride, groups, expansion, shortcut
|
158 |
+
super().__init__()
|
159 |
+
c_ = int(c2 * e) # hidden channels
|
160 |
+
self.cv1 = Conv(c1, c_, (1, k), (1, s))
|
161 |
+
self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g)
|
162 |
+
self.add = shortcut and c1 == c2
|
163 |
+
|
164 |
+
def forward(self, x):
|
165 |
+
return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
|
166 |
+
|
167 |
+
|
168 |
+
class C3(nn.Module):
|
169 |
+
# CSP Bottleneck with 3 convolutions
|
170 |
+
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5): # ch_in, ch_out, number, shortcut, groups, expansion
|
171 |
+
super().__init__()
|
172 |
+
c_ = int(c2 * e) # hidden channels
|
173 |
+
self.cv1 = Conv(c1, c_, 1, 1)
|
174 |
+
self.cv2 = Conv(c1, c_, 1, 1)
|
175 |
+
self.cv3 = Conv(2 * c_, c2, 1) # optional act=FReLU(c2)
|
176 |
+
self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))
|
177 |
+
|
178 |
+
def forward(self, x):
|
179 |
+
return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), 1))
|
180 |
+
|
181 |
+
|
182 |
+
class C3x(C3):
|
183 |
+
# C3 module with cross-convolutions
|
184 |
+
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
|
185 |
+
super().__init__(c1, c2, n, shortcut, g, e)
|
186 |
+
c_ = int(c2 * e)
|
187 |
+
self.m = nn.Sequential(*(CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)))
|
188 |
+
|
189 |
+
|
190 |
+
class C3TR(C3):
|
191 |
+
# C3 module with TransformerBlock()
|
192 |
+
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
|
193 |
+
super().__init__(c1, c2, n, shortcut, g, e)
|
194 |
+
c_ = int(c2 * e)
|
195 |
+
self.m = TransformerBlock(c_, c_, 4, n)
|
196 |
+
|
197 |
+
|
198 |
+
class C3SPP(C3):
|
199 |
+
# C3 module with SPP()
|
200 |
+
def __init__(self, c1, c2, k=(5, 9, 13), n=1, shortcut=True, g=1, e=0.5):
|
201 |
+
super().__init__(c1, c2, n, shortcut, g, e)
|
202 |
+
c_ = int(c2 * e)
|
203 |
+
self.m = SPP(c_, c_, k)
|
204 |
+
|
205 |
+
|
206 |
+
class C3Ghost(C3):
|
207 |
+
# C3 module with GhostBottleneck()
|
208 |
+
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
|
209 |
+
super().__init__(c1, c2, n, shortcut, g, e)
|
210 |
+
c_ = int(c2 * e) # hidden channels
|
211 |
+
self.m = nn.Sequential(*(GhostBottleneck(c_, c_) for _ in range(n)))
|
212 |
+
|
213 |
+
|
214 |
+
class SPP(nn.Module):
|
215 |
+
# Spatial Pyramid Pooling (SPP) layer https://arxiv.org/abs/1406.4729
|
216 |
+
def __init__(self, c1, c2, k=(5, 9, 13)):
|
217 |
+
super().__init__()
|
218 |
+
c_ = c1 // 2 # hidden channels
|
219 |
+
self.cv1 = Conv(c1, c_, 1, 1)
|
220 |
+
self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1)
|
221 |
+
self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k])
|
222 |
+
|
223 |
+
def forward(self, x):
|
224 |
+
x = self.cv1(x)
|
225 |
+
with warnings.catch_warnings():
|
226 |
+
warnings.simplefilter('ignore') # suppress torch 1.9.0 max_pool2d() warning
|
227 |
+
return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1))
|
228 |
+
|
229 |
+
|
230 |
+
class SPPF(nn.Module):
|
231 |
+
# Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher
|
232 |
+
def __init__(self, c1, c2, k=5): # equivalent to SPP(k=(5, 9, 13))
|
233 |
+
super().__init__()
|
234 |
+
c_ = c1 // 2 # hidden channels
|
235 |
+
self.cv1 = Conv(c1, c_, 1, 1)
|
236 |
+
self.cv2 = Conv(c_ * 4, c2, 1, 1)
|
237 |
+
self.m = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
|
238 |
+
|
239 |
+
def forward(self, x):
|
240 |
+
x = self.cv1(x)
|
241 |
+
with warnings.catch_warnings():
|
242 |
+
warnings.simplefilter('ignore') # suppress torch 1.9.0 max_pool2d() warning
|
243 |
+
y1 = self.m(x)
|
244 |
+
y2 = self.m(y1)
|
245 |
+
return self.cv2(torch.cat((x, y1, y2, self.m(y2)), 1))
|
246 |
+
|
247 |
+
|
248 |
+
class Focus(nn.Module):
|
249 |
+
# Focus wh information into c-space
|
250 |
+
def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True): # ch_in, ch_out, kernel, stride, padding, groups
|
251 |
+
super().__init__()
|
252 |
+
self.conv = Conv(c1 * 4, c2, k, s, p, g, act=act)
|
253 |
+
# self.contract = Contract(gain=2)
|
254 |
+
|
255 |
+
def forward(self, x): # x(b,c,w,h) -> y(b,4c,w/2,h/2)
|
256 |
+
return self.conv(torch.cat((x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]), 1))
|
257 |
+
# return self.conv(self.contract(x))
|
258 |
+
|
259 |
+
|
260 |
+
class GhostConv(nn.Module):
|
261 |
+
# Ghost Convolution https://github.com/huawei-noah/ghostnet
|
262 |
+
def __init__(self, c1, c2, k=1, s=1, g=1, act=True): # ch_in, ch_out, kernel, stride, groups
|
263 |
+
super().__init__()
|
264 |
+
c_ = c2 // 2 # hidden channels
|
265 |
+
self.cv1 = Conv(c1, c_, k, s, None, g, act=act)
|
266 |
+
self.cv2 = Conv(c_, c_, 5, 1, None, c_, act=act)
|
267 |
+
|
268 |
+
def forward(self, x):
|
269 |
+
y = self.cv1(x)
|
270 |
+
return torch.cat((y, self.cv2(y)), 1)
|
271 |
+
|
272 |
+
|
273 |
+
class GhostBottleneck(nn.Module):
|
274 |
+
# Ghost Bottleneck https://github.com/huawei-noah/ghostnet
|
275 |
+
def __init__(self, c1, c2, k=3, s=1): # ch_in, ch_out, kernel, stride
|
276 |
+
super().__init__()
|
277 |
+
c_ = c2 // 2
|
278 |
+
self.conv = nn.Sequential(
|
279 |
+
GhostConv(c1, c_, 1, 1), # pw
|
280 |
+
DWConv(c_, c_, k, s, act=False) if s == 2 else nn.Identity(), # dw
|
281 |
+
GhostConv(c_, c2, 1, 1, act=False)) # pw-linear
|
282 |
+
self.shortcut = nn.Sequential(DWConv(c1, c1, k, s, act=False), Conv(c1, c2, 1, 1,
|
283 |
+
act=False)) if s == 2 else nn.Identity()
|
284 |
+
|
285 |
+
def forward(self, x):
|
286 |
+
return self.conv(x) + self.shortcut(x)
|
287 |
+
|
288 |
+
|
289 |
+
class Contract(nn.Module):
|
290 |
+
# Contract width-height into channels, i.e. x(1,64,80,80) to x(1,256,40,40)
|
291 |
+
def __init__(self, gain=2):
|
292 |
+
super().__init__()
|
293 |
+
self.gain = gain
|
294 |
+
|
295 |
+
def forward(self, x):
|
296 |
+
b, c, h, w = x.size() # assert (h / s == 0) and (W / s == 0), 'Indivisible gain'
|
297 |
+
s = self.gain
|
298 |
+
x = x.view(b, c, h // s, s, w // s, s) # x(1,64,40,2,40,2)
|
299 |
+
x = x.permute(0, 3, 5, 1, 2, 4).contiguous() # x(1,2,2,64,40,40)
|
300 |
+
return x.view(b, c * s * s, h // s, w // s) # x(1,256,40,40)
|
301 |
+
|
302 |
+
|
303 |
+
class Expand(nn.Module):
|
304 |
+
# Expand channels into width-height, i.e. x(1,64,80,80) to x(1,16,160,160)
|
305 |
+
def __init__(self, gain=2):
|
306 |
+
super().__init__()
|
307 |
+
self.gain = gain
|
308 |
+
|
309 |
+
def forward(self, x):
|
310 |
+
b, c, h, w = x.size() # assert C / s ** 2 == 0, 'Indivisible gain'
|
311 |
+
s = self.gain
|
312 |
+
x = x.view(b, s, s, c // s ** 2, h, w) # x(1,2,2,16,80,80)
|
313 |
+
x = x.permute(0, 3, 4, 1, 5, 2).contiguous() # x(1,16,80,2,80,2)
|
314 |
+
return x.view(b, c // s ** 2, h * s, w * s) # x(1,16,160,160)
|
315 |
+
|
316 |
+
|
317 |
+
class Concat(nn.Module):
|
318 |
+
# Concatenate a list of tensors along dimension
|
319 |
+
def __init__(self, dimension=1):
|
320 |
+
super().__init__()
|
321 |
+
self.d = dimension
|
322 |
+
|
323 |
+
def forward(self, x):
|
324 |
+
return torch.cat(x, self.d)
|
325 |
+
|
326 |
+
|
327 |
+
class DetectMultiBackend(nn.Module):
|
328 |
+
# YOLOv5 MultiBackend class for python inference on various backends
|
329 |
+
def __init__(self, weights='yolov5s.pt', device=torch.device('cpu'), dnn=False, data=None, fp16=False, fuse=True):
|
330 |
+
# Usage:
|
331 |
+
# PyTorch: weights = *.pt
|
332 |
+
# TorchScript: *.torchscript
|
333 |
+
# ONNX Runtime: *.onnx
|
334 |
+
# ONNX OpenCV DNN: *.onnx --dnn
|
335 |
+
# OpenVINO: *_openvino_model
|
336 |
+
# CoreML: *.mlmodel
|
337 |
+
# TensorRT: *.engine
|
338 |
+
# TensorFlow SavedModel: *_saved_model
|
339 |
+
# TensorFlow GraphDef: *.pb
|
340 |
+
# TensorFlow Lite: *.tflite
|
341 |
+
# TensorFlow Edge TPU: *_edgetpu.tflite
|
342 |
+
# PaddlePaddle: *_paddle_model
|
343 |
+
from models.experimental import attempt_download, attempt_load # scoped to avoid circular import
|
344 |
+
|
345 |
+
super().__init__()
|
346 |
+
w = str(weights[0] if isinstance(weights, list) else weights)
|
347 |
+
pt, jit, onnx, xml, engine, coreml, saved_model, pb, tflite, edgetpu, tfjs, paddle, triton = self._model_type(w)
|
348 |
+
fp16 &= pt or jit or onnx or engine or triton # FP16
|
349 |
+
nhwc = coreml or saved_model or pb or tflite or edgetpu # BHWC formats (vs torch BCWH)
|
350 |
+
stride = 32 # default stride
|
351 |
+
cuda = torch.cuda.is_available() and device.type != 'cpu' # use CUDA
|
352 |
+
if not (pt or triton):
|
353 |
+
w = attempt_download(w) # download if not local
|
354 |
+
|
355 |
+
if pt: # PyTorch
|
356 |
+
model = attempt_load(weights if isinstance(weights, list) else w, device=device, inplace=True, fuse=fuse)
|
357 |
+
stride = max(int(model.stride.max()), 32) # model stride
|
358 |
+
names = model.module.names if hasattr(model, 'module') else model.names # get class names
|
359 |
+
model.half() if fp16 else model.float()
|
360 |
+
self.model = model # explicitly assign for to(), cpu(), cuda(), half()
|
361 |
+
elif jit: # TorchScript
|
362 |
+
LOGGER.info(f'Loading {w} for TorchScript inference...')
|
363 |
+
extra_files = {'config.txt': ''} # model metadata
|
364 |
+
model = torch.jit.load(w, _extra_files=extra_files, map_location=device)
|
365 |
+
model.half() if fp16 else model.float()
|
366 |
+
if extra_files['config.txt']: # load metadata dict
|
367 |
+
d = json.loads(extra_files['config.txt'],
|
368 |
+
object_hook=lambda d: {
|
369 |
+
int(k) if k.isdigit() else k: v
|
370 |
+
for k, v in d.items()})
|
371 |
+
stride, names = int(d['stride']), d['names']
|
372 |
+
elif dnn: # ONNX OpenCV DNN
|
373 |
+
LOGGER.info(f'Loading {w} for ONNX OpenCV DNN inference...')
|
374 |
+
check_requirements('opencv-python>=4.5.4')
|
375 |
+
net = cv2.dnn.readNetFromONNX(w)
|
376 |
+
elif onnx: # ONNX Runtime
|
377 |
+
LOGGER.info(f'Loading {w} for ONNX Runtime inference...')
|
378 |
+
check_requirements(('onnx', 'onnxruntime-gpu' if cuda else 'onnxruntime'))
|
379 |
+
import onnxruntime
|
380 |
+
providers = ['CUDAExecutionProvider', 'CPUExecutionProvider'] if cuda else ['CPUExecutionProvider']
|
381 |
+
session = onnxruntime.InferenceSession(w, providers=providers)
|
382 |
+
output_names = [x.name for x in session.get_outputs()]
|
383 |
+
meta = session.get_modelmeta().custom_metadata_map # metadata
|
384 |
+
if 'stride' in meta:
|
385 |
+
stride, names = int(meta['stride']), eval(meta['names'])
|
386 |
+
elif xml: # OpenVINO
|
387 |
+
LOGGER.info(f'Loading {w} for OpenVINO inference...')
|
388 |
+
check_requirements('openvino>=2023.0') # requires openvino-dev: https://pypi.org/project/openvino-dev/
|
389 |
+
from openvino.runtime import Core, Layout, get_batch
|
390 |
+
core = Core()
|
391 |
+
if not Path(w).is_file(): # if not *.xml
|
392 |
+
w = next(Path(w).glob('*.xml')) # get *.xml file from *_openvino_model dir
|
393 |
+
ov_model = core.read_model(model=w, weights=Path(w).with_suffix('.bin'))
|
394 |
+
if ov_model.get_parameters()[0].get_layout().empty:
|
395 |
+
ov_model.get_parameters()[0].set_layout(Layout('NCHW'))
|
396 |
+
batch_dim = get_batch(ov_model)
|
397 |
+
if batch_dim.is_static:
|
398 |
+
batch_size = batch_dim.get_length()
|
399 |
+
ov_compiled_model = core.compile_model(ov_model, device_name='AUTO') # AUTO selects best available device
|
400 |
+
stride, names = self._load_metadata(Path(w).with_suffix('.yaml')) # load metadata
|
401 |
+
elif engine: # TensorRT
|
402 |
+
LOGGER.info(f'Loading {w} for TensorRT inference...')
|
403 |
+
import tensorrt as trt # https://developer.nvidia.com/nvidia-tensorrt-download
|
404 |
+
check_version(trt.__version__, '7.0.0', hard=True) # require tensorrt>=7.0.0
|
405 |
+
if device.type == 'cpu':
|
406 |
+
device = torch.device('cuda:0')
|
407 |
+
Binding = namedtuple('Binding', ('name', 'dtype', 'shape', 'data', 'ptr'))
|
408 |
+
logger = trt.Logger(trt.Logger.INFO)
|
409 |
+
with open(w, 'rb') as f, trt.Runtime(logger) as runtime:
|
410 |
+
model = runtime.deserialize_cuda_engine(f.read())
|
411 |
+
context = model.create_execution_context()
|
412 |
+
bindings = OrderedDict()
|
413 |
+
output_names = []
|
414 |
+
fp16 = False # default updated below
|
415 |
+
dynamic = False
|
416 |
+
for i in range(model.num_bindings):
|
417 |
+
name = model.get_binding_name(i)
|
418 |
+
dtype = trt.nptype(model.get_binding_dtype(i))
|
419 |
+
if model.binding_is_input(i):
|
420 |
+
if -1 in tuple(model.get_binding_shape(i)): # dynamic
|
421 |
+
dynamic = True
|
422 |
+
context.set_binding_shape(i, tuple(model.get_profile_shape(0, i)[2]))
|
423 |
+
if dtype == np.float16:
|
424 |
+
fp16 = True
|
425 |
+
else: # output
|
426 |
+
output_names.append(name)
|
427 |
+
shape = tuple(context.get_binding_shape(i))
|
428 |
+
im = torch.from_numpy(np.empty(shape, dtype=dtype)).to(device)
|
429 |
+
bindings[name] = Binding(name, dtype, shape, im, int(im.data_ptr()))
|
430 |
+
binding_addrs = OrderedDict((n, d.ptr) for n, d in bindings.items())
|
431 |
+
batch_size = bindings['images'].shape[0] # if dynamic, this is instead max batch size
|
432 |
+
elif coreml: # CoreML
|
433 |
+
LOGGER.info(f'Loading {w} for CoreML inference...')
|
434 |
+
import coremltools as ct
|
435 |
+
model = ct.models.MLModel(w)
|
436 |
+
elif saved_model: # TF SavedModel
|
437 |
+
LOGGER.info(f'Loading {w} for TensorFlow SavedModel inference...')
|
438 |
+
import tensorflow as tf
|
439 |
+
keras = False # assume TF1 saved_model
|
440 |
+
model = tf.keras.models.load_model(w) if keras else tf.saved_model.load(w)
|
441 |
+
elif pb: # GraphDef https://www.tensorflow.org/guide/migrate#a_graphpb_or_graphpbtxt
|
442 |
+
LOGGER.info(f'Loading {w} for TensorFlow GraphDef inference...')
|
443 |
+
import tensorflow as tf
|
444 |
+
|
445 |
+
def wrap_frozen_graph(gd, inputs, outputs):
|
446 |
+
x = tf.compat.v1.wrap_function(lambda: tf.compat.v1.import_graph_def(gd, name=''), []) # wrapped
|
447 |
+
ge = x.graph.as_graph_element
|
448 |
+
return x.prune(tf.nest.map_structure(ge, inputs), tf.nest.map_structure(ge, outputs))
|
449 |
+
|
450 |
+
def gd_outputs(gd):
|
451 |
+
name_list, input_list = [], []
|
452 |
+
for node in gd.node: # tensorflow.core.framework.node_def_pb2.NodeDef
|
453 |
+
name_list.append(node.name)
|
454 |
+
input_list.extend(node.input)
|
455 |
+
return sorted(f'{x}:0' for x in list(set(name_list) - set(input_list)) if not x.startswith('NoOp'))
|
456 |
+
|
457 |
+
gd = tf.Graph().as_graph_def() # TF GraphDef
|
458 |
+
with open(w, 'rb') as f:
|
459 |
+
gd.ParseFromString(f.read())
|
460 |
+
frozen_func = wrap_frozen_graph(gd, inputs='x:0', outputs=gd_outputs(gd))
|
461 |
+
elif tflite or edgetpu: # https://www.tensorflow.org/lite/guide/python#install_tensorflow_lite_for_python
|
462 |
+
try: # https://coral.ai/docs/edgetpu/tflite-python/#update-existing-tf-lite-code-for-the-edge-tpu
|
463 |
+
from tflite_runtime.interpreter import Interpreter, load_delegate
|
464 |
+
except ImportError:
|
465 |
+
import tensorflow as tf
|
466 |
+
Interpreter, load_delegate = tf.lite.Interpreter, tf.lite.experimental.load_delegate,
|
467 |
+
if edgetpu: # TF Edge TPU https://coral.ai/software/#edgetpu-runtime
|
468 |
+
LOGGER.info(f'Loading {w} for TensorFlow Lite Edge TPU inference...')
|
469 |
+
delegate = {
|
470 |
+
'Linux': 'libedgetpu.so.1',
|
471 |
+
'Darwin': 'libedgetpu.1.dylib',
|
472 |
+
'Windows': 'edgetpu.dll'}[platform.system()]
|
473 |
+
interpreter = Interpreter(model_path=w, experimental_delegates=[load_delegate(delegate)])
|
474 |
+
else: # TFLite
|
475 |
+
LOGGER.info(f'Loading {w} for TensorFlow Lite inference...')
|
476 |
+
interpreter = Interpreter(model_path=w) # load TFLite model
|
477 |
+
interpreter.allocate_tensors() # allocate
|
478 |
+
input_details = interpreter.get_input_details() # inputs
|
479 |
+
output_details = interpreter.get_output_details() # outputs
|
480 |
+
# load metadata
|
481 |
+
with contextlib.suppress(zipfile.BadZipFile):
|
482 |
+
with zipfile.ZipFile(w, 'r') as model:
|
483 |
+
meta_file = model.namelist()[0]
|
484 |
+
meta = ast.literal_eval(model.read(meta_file).decode('utf-8'))
|
485 |
+
stride, names = int(meta['stride']), meta['names']
|
486 |
+
elif tfjs: # TF.js
|
487 |
+
raise NotImplementedError('ERROR: YOLOv5 TF.js inference is not supported')
|
488 |
+
elif paddle: # PaddlePaddle
|
489 |
+
LOGGER.info(f'Loading {w} for PaddlePaddle inference...')
|
490 |
+
check_requirements('paddlepaddle-gpu' if cuda else 'paddlepaddle')
|
491 |
+
import paddle.inference as pdi
|
492 |
+
if not Path(w).is_file(): # if not *.pdmodel
|
493 |
+
w = next(Path(w).rglob('*.pdmodel')) # get *.pdmodel file from *_paddle_model dir
|
494 |
+
weights = Path(w).with_suffix('.pdiparams')
|
495 |
+
config = pdi.Config(str(w), str(weights))
|
496 |
+
if cuda:
|
497 |
+
config.enable_use_gpu(memory_pool_init_size_mb=2048, device_id=0)
|
498 |
+
predictor = pdi.create_predictor(config)
|
499 |
+
input_handle = predictor.get_input_handle(predictor.get_input_names()[0])
|
500 |
+
output_names = predictor.get_output_names()
|
501 |
+
elif triton: # NVIDIA Triton Inference Server
|
502 |
+
LOGGER.info(f'Using {w} as Triton Inference Server...')
|
503 |
+
check_requirements('tritonclient[all]')
|
504 |
+
from utils.triton import TritonRemoteModel
|
505 |
+
model = TritonRemoteModel(url=w)
|
506 |
+
nhwc = model.runtime.startswith('tensorflow')
|
507 |
+
else:
|
508 |
+
raise NotImplementedError(f'ERROR: {w} is not a supported format')
|
509 |
+
|
510 |
+
# class names
|
511 |
+
if 'names' not in locals():
|
512 |
+
names = yaml_load(data)['names'] if data else {i: f'class{i}' for i in range(999)}
|
513 |
+
if names[0] == 'n01440764' and len(names) == 1000: # ImageNet
|
514 |
+
names = yaml_load(ROOT / 'data/ImageNet.yaml')['names'] # human-readable names
|
515 |
+
|
516 |
+
self.__dict__.update(locals()) # assign all variables to self
|
517 |
+
|
518 |
+
def forward(self, im, augment=False, visualize=False):
|
519 |
+
# YOLOv5 MultiBackend inference
|
520 |
+
b, ch, h, w = im.shape # batch, channel, height, width
|
521 |
+
if self.fp16 and im.dtype != torch.float16:
|
522 |
+
im = im.half() # to FP16
|
523 |
+
if self.nhwc:
|
524 |
+
im = im.permute(0, 2, 3, 1) # torch BCHW to numpy BHWC shape(1,320,192,3)
|
525 |
+
|
526 |
+
if self.pt: # PyTorch
|
527 |
+
y = self.model(im, augment=augment, visualize=visualize) if augment or visualize else self.model(im)
|
528 |
+
elif self.jit: # TorchScript
|
529 |
+
y = self.model(im)
|
530 |
+
elif self.dnn: # ONNX OpenCV DNN
|
531 |
+
im = im.cpu().numpy() # torch to numpy
|
532 |
+
self.net.setInput(im)
|
533 |
+
y = self.net.forward()
|
534 |
+
elif self.onnx: # ONNX Runtime
|
535 |
+
im = im.cpu().numpy() # torch to numpy
|
536 |
+
y = self.session.run(self.output_names, {self.session.get_inputs()[0].name: im})
|
537 |
+
elif self.xml: # OpenVINO
|
538 |
+
im = im.cpu().numpy() # FP32
|
539 |
+
y = list(self.ov_compiled_model(im).values())
|
540 |
+
elif self.engine: # TensorRT
|
541 |
+
if self.dynamic and im.shape != self.bindings['images'].shape:
|
542 |
+
i = self.model.get_binding_index('images')
|
543 |
+
self.context.set_binding_shape(i, im.shape) # reshape if dynamic
|
544 |
+
self.bindings['images'] = self.bindings['images']._replace(shape=im.shape)
|
545 |
+
for name in self.output_names:
|
546 |
+
i = self.model.get_binding_index(name)
|
547 |
+
self.bindings[name].data.resize_(tuple(self.context.get_binding_shape(i)))
|
548 |
+
s = self.bindings['images'].shape
|
549 |
+
assert im.shape == s, f"input size {im.shape} {'>' if self.dynamic else 'not equal to'} max model size {s}"
|
550 |
+
self.binding_addrs['images'] = int(im.data_ptr())
|
551 |
+
self.context.execute_v2(list(self.binding_addrs.values()))
|
552 |
+
y = [self.bindings[x].data for x in sorted(self.output_names)]
|
553 |
+
elif self.coreml: # CoreML
|
554 |
+
im = im.cpu().numpy()
|
555 |
+
im = Image.fromarray((im[0] * 255).astype('uint8'))
|
556 |
+
# im = im.resize((192, 320), Image.BILINEAR)
|
557 |
+
y = self.model.predict({'image': im}) # coordinates are xywh normalized
|
558 |
+
if 'confidence' in y:
|
559 |
+
box = xywh2xyxy(y['coordinates'] * [[w, h, w, h]]) # xyxy pixels
|
560 |
+
conf, cls = y['confidence'].max(1), y['confidence'].argmax(1).astype(np.float)
|
561 |
+
y = np.concatenate((box, conf.reshape(-1, 1), cls.reshape(-1, 1)), 1)
|
562 |
+
else:
|
563 |
+
y = list(reversed(y.values())) # reversed for segmentation models (pred, proto)
|
564 |
+
elif self.paddle: # PaddlePaddle
|
565 |
+
im = im.cpu().numpy().astype(np.float32)
|
566 |
+
self.input_handle.copy_from_cpu(im)
|
567 |
+
self.predictor.run()
|
568 |
+
y = [self.predictor.get_output_handle(x).copy_to_cpu() for x in self.output_names]
|
569 |
+
elif self.triton: # NVIDIA Triton Inference Server
|
570 |
+
y = self.model(im)
|
571 |
+
else: # TensorFlow (SavedModel, GraphDef, Lite, Edge TPU)
|
572 |
+
im = im.cpu().numpy()
|
573 |
+
if self.saved_model: # SavedModel
|
574 |
+
y = self.model(im, training=False) if self.keras else self.model(im)
|
575 |
+
elif self.pb: # GraphDef
|
576 |
+
y = self.frozen_func(x=self.tf.constant(im))
|
577 |
+
else: # Lite or Edge TPU
|
578 |
+
input = self.input_details[0]
|
579 |
+
int8 = input['dtype'] == np.uint8 # is TFLite quantized uint8 model
|
580 |
+
if int8:
|
581 |
+
scale, zero_point = input['quantization']
|
582 |
+
im = (im / scale + zero_point).astype(np.uint8) # de-scale
|
583 |
+
self.interpreter.set_tensor(input['index'], im)
|
584 |
+
self.interpreter.invoke()
|
585 |
+
y = []
|
586 |
+
for output in self.output_details:
|
587 |
+
x = self.interpreter.get_tensor(output['index'])
|
588 |
+
if int8:
|
589 |
+
scale, zero_point = output['quantization']
|
590 |
+
x = (x.astype(np.float32) - zero_point) * scale # re-scale
|
591 |
+
y.append(x)
|
592 |
+
y = [x if isinstance(x, np.ndarray) else x.numpy() for x in y]
|
593 |
+
y[0][..., :4] *= [w, h, w, h] # xywh normalized to pixels
|
594 |
+
|
595 |
+
if isinstance(y, (list, tuple)):
|
596 |
+
return self.from_numpy(y[0]) if len(y) == 1 else [self.from_numpy(x) for x in y]
|
597 |
+
else:
|
598 |
+
return self.from_numpy(y)
|
599 |
+
|
600 |
+
def from_numpy(self, x):
|
601 |
+
return torch.from_numpy(x).to(self.device) if isinstance(x, np.ndarray) else x
|
602 |
+
|
603 |
+
def warmup(self, imgsz=(1, 3, 640, 640)):
|
604 |
+
# Warmup model by running inference once
|
605 |
+
warmup_types = self.pt, self.jit, self.onnx, self.engine, self.saved_model, self.pb, self.triton
|
606 |
+
if any(warmup_types) and (self.device.type != 'cpu' or self.triton):
|
607 |
+
im = torch.empty(*imgsz, dtype=torch.half if self.fp16 else torch.float, device=self.device) # input
|
608 |
+
for _ in range(2 if self.jit else 1): #
|
609 |
+
self.forward(im) # warmup
|
610 |
+
|
611 |
+
@staticmethod
|
612 |
+
def _model_type(p='path/to/model.pt'):
|
613 |
+
# Return model type from model path, i.e. path='path/to/model.onnx' -> type=onnx
|
614 |
+
# types = [pt, jit, onnx, xml, engine, coreml, saved_model, pb, tflite, edgetpu, tfjs, paddle]
|
615 |
+
from export import export_formats
|
616 |
+
from utils.downloads import is_url
|
617 |
+
sf = list(export_formats().Suffix) # export suffixes
|
618 |
+
if not is_url(p, check=False):
|
619 |
+
check_suffix(p, sf) # checks
|
620 |
+
url = urlparse(p) # if url may be Triton inference server
|
621 |
+
types = [s in Path(p).name for s in sf]
|
622 |
+
types[8] &= not types[9] # tflite &= not edgetpu
|
623 |
+
triton = not any(types) and all([any(s in url.scheme for s in ['http', 'grpc']), url.netloc])
|
624 |
+
return types + [triton]
|
625 |
+
|
626 |
+
@staticmethod
|
627 |
+
def _load_metadata(f=Path('path/to/meta.yaml')):
|
628 |
+
# Load metadata from meta.yaml if it exists
|
629 |
+
if f.exists():
|
630 |
+
d = yaml_load(f)
|
631 |
+
return d['stride'], d['names'] # assign stride, names
|
632 |
+
return None, None
|
633 |
+
|
634 |
+
|
635 |
+
class AutoShape(nn.Module):
|
636 |
+
# YOLOv5 input-robust model wrapper for passing cv2/np/PIL/torch inputs. Includes preprocessing, inference and NMS
|
637 |
+
conf = 0.25 # NMS confidence threshold
|
638 |
+
iou = 0.45 # NMS IoU threshold
|
639 |
+
agnostic = False # NMS class-agnostic
|
640 |
+
multi_label = False # NMS multiple labels per box
|
641 |
+
classes = None # (optional list) filter by class, i.e. = [0, 15, 16] for COCO persons, cats and dogs
|
642 |
+
max_det = 1000 # maximum number of detections per image
|
643 |
+
amp = False # Automatic Mixed Precision (AMP) inference
|
644 |
+
|
645 |
+
def __init__(self, model, verbose=True):
|
646 |
+
super().__init__()
|
647 |
+
if verbose:
|
648 |
+
LOGGER.info('Adding AutoShape... ')
|
649 |
+
copy_attr(self, model, include=('yaml', 'nc', 'hyp', 'names', 'stride', 'abc'), exclude=()) # copy attributes
|
650 |
+
self.dmb = isinstance(model, DetectMultiBackend) # DetectMultiBackend() instance
|
651 |
+
self.pt = not self.dmb or model.pt # PyTorch model
|
652 |
+
self.model = model.eval()
|
653 |
+
if self.pt:
|
654 |
+
m = self.model.model.model[-1] if self.dmb else self.model.model[-1] # Detect()
|
655 |
+
m.inplace = False # Detect.inplace=False for safe multithread inference
|
656 |
+
m.export = True # do not output loss values
|
657 |
+
|
658 |
+
def _apply(self, fn):
|
659 |
+
# Apply to(), cpu(), cuda(), half() to model tensors that are not parameters or registered buffers
|
660 |
+
self = super()._apply(fn)
|
661 |
+
if self.pt:
|
662 |
+
m = self.model.model.model[-1] if self.dmb else self.model.model[-1] # Detect()
|
663 |
+
m.stride = fn(m.stride)
|
664 |
+
m.grid = list(map(fn, m.grid))
|
665 |
+
if isinstance(m.anchor_grid, list):
|
666 |
+
m.anchor_grid = list(map(fn, m.anchor_grid))
|
667 |
+
return self
|
668 |
+
|
669 |
+
@smart_inference_mode()
|
670 |
+
def forward(self, ims, size=640, augment=False, profile=False):
|
671 |
+
# Inference from various sources. For size(height=640, width=1280), RGB images example inputs are:
|
672 |
+
# file: ims = 'data/images/zidane.jpg' # str or PosixPath
|
673 |
+
# URI: = 'https://ultralytics.com/images/zidane.jpg'
|
674 |
+
# OpenCV: = cv2.imread('image.jpg')[:,:,::-1] # HWC BGR to RGB x(640,1280,3)
|
675 |
+
# PIL: = Image.open('image.jpg') or ImageGrab.grab() # HWC x(640,1280,3)
|
676 |
+
# numpy: = np.zeros((640,1280,3)) # HWC
|
677 |
+
# torch: = torch.zeros(16,3,320,640) # BCHW (scaled to size=640, 0-1 values)
|
678 |
+
# multiple: = [Image.open('image1.jpg'), Image.open('image2.jpg'), ...] # list of images
|
679 |
+
|
680 |
+
dt = (Profile(), Profile(), Profile())
|
681 |
+
with dt[0]:
|
682 |
+
if isinstance(size, int): # expand
|
683 |
+
size = (size, size)
|
684 |
+
p = next(self.model.parameters()) if self.pt else torch.empty(1, device=self.model.device) # param
|
685 |
+
autocast = self.amp and (p.device.type != 'cpu') # Automatic Mixed Precision (AMP) inference
|
686 |
+
if isinstance(ims, torch.Tensor): # torch
|
687 |
+
with amp.autocast(autocast):
|
688 |
+
return self.model(ims.to(p.device).type_as(p), augment=augment) # inference
|
689 |
+
|
690 |
+
# Pre-process
|
691 |
+
n, ims = (len(ims), list(ims)) if isinstance(ims, (list, tuple)) else (1, [ims]) # number, list of images
|
692 |
+
shape0, shape1, files = [], [], [] # image and inference shapes, filenames
|
693 |
+
for i, im in enumerate(ims):
|
694 |
+
f = f'image{i}' # filename
|
695 |
+
if isinstance(im, (str, Path)): # filename or uri
|
696 |
+
im, f = Image.open(requests.get(im, stream=True).raw if str(im).startswith('http') else im), im
|
697 |
+
im = np.asarray(exif_transpose(im))
|
698 |
+
elif isinstance(im, Image.Image): # PIL Image
|
699 |
+
im, f = np.asarray(exif_transpose(im)), getattr(im, 'filename', f) or f
|
700 |
+
files.append(Path(f).with_suffix('.jpg').name)
|
701 |
+
if im.shape[0] < 5: # image in CHW
|
702 |
+
im = im.transpose((1, 2, 0)) # reverse dataloader .transpose(2, 0, 1)
|
703 |
+
im = im[..., :3] if im.ndim == 3 else cv2.cvtColor(im, cv2.COLOR_GRAY2BGR) # enforce 3ch input
|
704 |
+
s = im.shape[:2] # HWC
|
705 |
+
shape0.append(s) # image shape
|
706 |
+
g = max(size) / max(s) # gain
|
707 |
+
shape1.append([int(y * g) for y in s])
|
708 |
+
ims[i] = im if im.data.contiguous else np.ascontiguousarray(im) # update
|
709 |
+
shape1 = [make_divisible(x, self.stride) for x in np.array(shape1).max(0)] # inf shape
|
710 |
+
x = [letterbox(im, shape1, auto=False)[0] for im in ims] # pad
|
711 |
+
x = np.ascontiguousarray(np.array(x).transpose((0, 3, 1, 2))) # stack and BHWC to BCHW
|
712 |
+
x = torch.from_numpy(x).to(p.device).type_as(p) / 255 # uint8 to fp16/32
|
713 |
+
|
714 |
+
with amp.autocast(autocast):
|
715 |
+
# Inference
|
716 |
+
with dt[1]:
|
717 |
+
y = self.model(x, augment=augment) # forward
|
718 |
+
|
719 |
+
# Post-process
|
720 |
+
with dt[2]:
|
721 |
+
y = non_max_suppression(y if self.dmb else y[0],
|
722 |
+
self.conf,
|
723 |
+
self.iou,
|
724 |
+
self.classes,
|
725 |
+
self.agnostic,
|
726 |
+
self.multi_label,
|
727 |
+
max_det=self.max_det) # NMS
|
728 |
+
for i in range(n):
|
729 |
+
scale_boxes(shape1, y[i][:, :4], shape0[i])
|
730 |
+
|
731 |
+
return Detections(ims, y, files, dt, self.names, x.shape)
|
732 |
+
|
733 |
+
|
734 |
+
class Detections:
|
735 |
+
# YOLOv5 detections class for inference results
|
736 |
+
def __init__(self, ims, pred, files, times=(0, 0, 0), names=None, shape=None):
|
737 |
+
super().__init__()
|
738 |
+
d = pred[0].device # device
|
739 |
+
gn = [torch.tensor([*(im.shape[i] for i in [1, 0, 1, 0]), 1, 1], device=d) for im in ims] # normalizations
|
740 |
+
self.ims = ims # list of images as numpy arrays
|
741 |
+
self.pred = pred # list of tensors pred[0] = (xyxy, conf, cls)
|
742 |
+
self.names = names # class names
|
743 |
+
self.files = files # image filenames
|
744 |
+
self.times = times # profiling times
|
745 |
+
self.xyxy = pred # xyxy pixels
|
746 |
+
self.xywh = [xyxy2xywh(x) for x in pred] # xywh pixels
|
747 |
+
self.xyxyn = [x / g for x, g in zip(self.xyxy, gn)] # xyxy normalized
|
748 |
+
self.xywhn = [x / g for x, g in zip(self.xywh, gn)] # xywh normalized
|
749 |
+
self.n = len(self.pred) # number of images (batch size)
|
750 |
+
self.t = tuple(x.t / self.n * 1E3 for x in times) # timestamps (ms)
|
751 |
+
self.s = tuple(shape) # inference BCHW shape
|
752 |
+
|
753 |
+
def _run(self, pprint=False, show=False, save=False, crop=False, render=False, labels=True, save_dir=Path('')):
|
754 |
+
s, crops = '', []
|
755 |
+
for i, (im, pred) in enumerate(zip(self.ims, self.pred)):
|
756 |
+
s += f'\nimage {i + 1}/{len(self.pred)}: {im.shape[0]}x{im.shape[1]} ' # string
|
757 |
+
if pred.shape[0]:
|
758 |
+
for c in pred[:, -1].unique():
|
759 |
+
n = (pred[:, -1] == c).sum() # detections per class
|
760 |
+
s += f"{n} {self.names[int(c)]}{'s' * (n > 1)}, " # add to string
|
761 |
+
s = s.rstrip(', ')
|
762 |
+
if show or save or render or crop:
|
763 |
+
annotator = Annotator(im, example=str(self.names))
|
764 |
+
for *box, conf, cls in reversed(pred): # xyxy, confidence, class
|
765 |
+
label = f'{self.names[int(cls)]} {conf:.2f}'
|
766 |
+
if crop:
|
767 |
+
file = save_dir / 'crops' / self.names[int(cls)] / self.files[i] if save else None
|
768 |
+
crops.append({
|
769 |
+
'box': box,
|
770 |
+
'conf': conf,
|
771 |
+
'cls': cls,
|
772 |
+
'label': label,
|
773 |
+
'im': save_one_box(box, im, file=file, save=save)})
|
774 |
+
else: # all others
|
775 |
+
annotator.box_label(box, label if labels else '', color=colors(cls))
|
776 |
+
im = annotator.im
|
777 |
+
else:
|
778 |
+
s += '(no detections)'
|
779 |
+
|
780 |
+
im = Image.fromarray(im.astype(np.uint8)) if isinstance(im, np.ndarray) else im # from np
|
781 |
+
if show:
|
782 |
+
if is_jupyter():
|
783 |
+
from IPython.display import display
|
784 |
+
display(im)
|
785 |
+
else:
|
786 |
+
im.show(self.files[i])
|
787 |
+
if save:
|
788 |
+
f = self.files[i]
|
789 |
+
im.save(save_dir / f) # save
|
790 |
+
if i == self.n - 1:
|
791 |
+
LOGGER.info(f"Saved {self.n} image{'s' * (self.n > 1)} to {colorstr('bold', save_dir)}")
|
792 |
+
if render:
|
793 |
+
self.ims[i] = np.asarray(im)
|
794 |
+
if pprint:
|
795 |
+
s = s.lstrip('\n')
|
796 |
+
return f'{s}\nSpeed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {self.s}' % self.t
|
797 |
+
if crop:
|
798 |
+
if save:
|
799 |
+
LOGGER.info(f'Saved results to {save_dir}\n')
|
800 |
+
return crops
|
801 |
+
|
802 |
+
@TryExcept('Showing images is not supported in this environment')
|
803 |
+
def show(self, labels=True):
|
804 |
+
self._run(show=True, labels=labels) # show results
|
805 |
+
|
806 |
+
def save(self, labels=True, save_dir='runs/detect/exp', exist_ok=False):
|
807 |
+
save_dir = increment_path(save_dir, exist_ok, mkdir=True) # increment save_dir
|
808 |
+
self._run(save=True, labels=labels, save_dir=save_dir) # save results
|
809 |
+
|
810 |
+
def crop(self, save=True, save_dir='runs/detect/exp', exist_ok=False):
|
811 |
+
save_dir = increment_path(save_dir, exist_ok, mkdir=True) if save else None
|
812 |
+
return self._run(crop=True, save=save, save_dir=save_dir) # crop results
|
813 |
+
|
814 |
+
def render(self, labels=True):
|
815 |
+
self._run(render=True, labels=labels) # render results
|
816 |
+
return self.ims
|
817 |
+
|
818 |
+
def pandas(self):
|
819 |
+
# return detections as pandas DataFrames, i.e. print(results.pandas().xyxy[0])
|
820 |
+
new = copy(self) # return copy
|
821 |
+
ca = 'xmin', 'ymin', 'xmax', 'ymax', 'confidence', 'class', 'name' # xyxy columns
|
822 |
+
cb = 'xcenter', 'ycenter', 'width', 'height', 'confidence', 'class', 'name' # xywh columns
|
823 |
+
for k, c in zip(['xyxy', 'xyxyn', 'xywh', 'xywhn'], [ca, ca, cb, cb]):
|
824 |
+
a = [[x[:5] + [int(x[5]), self.names[int(x[5])]] for x in x.tolist()] for x in getattr(self, k)] # update
|
825 |
+
setattr(new, k, [pd.DataFrame(x, columns=c) for x in a])
|
826 |
+
return new
|
827 |
+
|
828 |
+
def tolist(self):
|
829 |
+
# return a list of Detections objects, i.e. 'for result in results.tolist():'
|
830 |
+
r = range(self.n) # iterable
|
831 |
+
x = [Detections([self.ims[i]], [self.pred[i]], [self.files[i]], self.times, self.names, self.s) for i in r]
|
832 |
+
# for d in x:
|
833 |
+
# for k in ['ims', 'pred', 'xyxy', 'xyxyn', 'xywh', 'xywhn']:
|
834 |
+
# setattr(d, k, getattr(d, k)[0]) # pop out of list
|
835 |
+
return x
|
836 |
+
|
837 |
+
def print(self):
|
838 |
+
LOGGER.info(self.__str__())
|
839 |
+
|
840 |
+
def __len__(self): # override len(results)
|
841 |
+
return self.n
|
842 |
+
|
843 |
+
def __str__(self): # override print(results)
|
844 |
+
return self._run(pprint=True) # print results
|
845 |
+
|
846 |
+
def __repr__(self):
|
847 |
+
return f'YOLOv5 {self.__class__} instance\n' + self.__str__()
|
848 |
+
|
849 |
+
|
850 |
+
class Proto(nn.Module):
|
851 |
+
# YOLOv5 mask Proto module for segmentation models
|
852 |
+
def __init__(self, c1, c_=256, c2=32): # ch_in, number of protos, number of masks
|
853 |
+
super().__init__()
|
854 |
+
self.cv1 = Conv(c1, c_, k=3)
|
855 |
+
self.upsample = nn.Upsample(scale_factor=2, mode='nearest')
|
856 |
+
self.cv2 = Conv(c_, c_, k=3)
|
857 |
+
self.cv3 = Conv(c_, c2)
|
858 |
+
|
859 |
+
def forward(self, x):
|
860 |
+
return self.cv3(self.cv2(self.upsample(self.cv1(x))))
|
861 |
+
|
862 |
+
|
863 |
+
class Classify(nn.Module):
|
864 |
+
# YOLOv5 classification head, i.e. x(b,c1,20,20) to x(b,c2)
|
865 |
+
def __init__(self,
|
866 |
+
c1,
|
867 |
+
c2,
|
868 |
+
k=1,
|
869 |
+
s=1,
|
870 |
+
p=None,
|
871 |
+
g=1,
|
872 |
+
dropout_p=0.0): # ch_in, ch_out, kernel, stride, padding, groups, dropout probability
|
873 |
+
super().__init__()
|
874 |
+
c_ = 1280 # efficientnet_b0 size
|
875 |
+
self.conv = Conv(c1, c_, k, s, autopad(k, p), g)
|
876 |
+
self.pool = nn.AdaptiveAvgPool2d(1) # to x(b,c_,1,1)
|
877 |
+
self.drop = nn.Dropout(p=dropout_p, inplace=True)
|
878 |
+
self.linear = nn.Linear(c_, c2) # to x(b,c2)
|
879 |
+
|
880 |
+
def forward(self, x):
|
881 |
+
if isinstance(x, list):
|
882 |
+
x = torch.cat(x, 1)
|
883 |
+
return self.linear(self.drop(self.pool(self.conv(x)).flatten(1)))
|
884 |
+
|
885 |
+
|
886 |
+
class CAM(nn.Module):
|
887 |
+
def __init__(self, inc, fusion='weight'):
|
888 |
+
super().__init__()
|
889 |
+
|
890 |
+
assert fusion in ['weight', 'adaptive', 'concat'] # δΈη§θεζΉεΌ
|
891 |
+
self.fusion = fusion
|
892 |
+
|
893 |
+
self.conv1 = Conv(inc, inc, 3, 1, None, 1, 1)
|
894 |
+
self.conv2 = Conv(inc, inc, 3, 1, None, 1, 3)
|
895 |
+
self.conv3 = Conv(inc, inc, 3, 1, None, 1, 5)
|
896 |
+
|
897 |
+
self.fusion_1 = Conv(inc, inc, 1)
|
898 |
+
self.fusion_2 = Conv(inc, inc, 1)
|
899 |
+
self.fusion_3 = Conv(inc, inc, 1)
|
900 |
+
|
901 |
+
if self.fusion == 'adaptive':
|
902 |
+
self.fusion_4 = Conv(inc * 3, 3, 1)
|
903 |
+
|
904 |
+
def forward(self, x):
|
905 |
+
x1 = self.conv1(x)
|
906 |
+
x2 = self.conv2(x)
|
907 |
+
x3 = self.conv3(x)
|
908 |
+
|
909 |
+
if self.fusion == 'weight':
|
910 |
+
return self.fusion_1(x1) + self.fusion_2(x2) + self.fusion_3(x3)
|
911 |
+
elif self.fusion == 'adaptive':
|
912 |
+
fusion = torch.softmax(
|
913 |
+
self.fusion_4(torch.cat([self.fusion_1(x1), self.fusion_2(x2), self.fusion_3(x3)], dim=1)), dim=1)
|
914 |
+
x1_weight, x2_weight, x3_weight = torch.split(fusion, [1, 1, 1], dim=1)
|
915 |
+
return x1 * x1_weight + x2 * x2_weight + x3 * x3_weight
|
916 |
+
else:
|
917 |
+
return torch.cat([self.fusion_1(x1), self.fusion_2(x2), self.fusion_3(x3)], dim=1)
|
918 |
+
|
models/experimental.py
ADDED
@@ -0,0 +1,111 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
"""
|
3 |
+
Experimental modules
|
4 |
+
"""
|
5 |
+
import math
|
6 |
+
|
7 |
+
import numpy as np
|
8 |
+
import torch
|
9 |
+
import torch.nn as nn
|
10 |
+
|
11 |
+
from utils.downloads import attempt_download
|
12 |
+
|
13 |
+
|
14 |
+
class Sum(nn.Module):
|
15 |
+
# Weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070
|
16 |
+
def __init__(self, n, weight=False): # n: number of inputs
|
17 |
+
super().__init__()
|
18 |
+
self.weight = weight # apply weights boolean
|
19 |
+
self.iter = range(n - 1) # iter object
|
20 |
+
if weight:
|
21 |
+
self.w = nn.Parameter(-torch.arange(1.0, n) / 2, requires_grad=True) # layer weights
|
22 |
+
|
23 |
+
def forward(self, x):
|
24 |
+
y = x[0] # no weight
|
25 |
+
if self.weight:
|
26 |
+
w = torch.sigmoid(self.w) * 2
|
27 |
+
for i in self.iter:
|
28 |
+
y = y + x[i + 1] * w[i]
|
29 |
+
else:
|
30 |
+
for i in self.iter:
|
31 |
+
y = y + x[i + 1]
|
32 |
+
return y
|
33 |
+
|
34 |
+
|
35 |
+
class MixConv2d(nn.Module):
|
36 |
+
# Mixed Depth-wise Conv https://arxiv.org/abs/1907.09595
|
37 |
+
def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True): # ch_in, ch_out, kernel, stride, ch_strategy
|
38 |
+
super().__init__()
|
39 |
+
n = len(k) # number of convolutions
|
40 |
+
if equal_ch: # equal c_ per group
|
41 |
+
i = torch.linspace(0, n - 1E-6, c2).floor() # c2 indices
|
42 |
+
c_ = [(i == g).sum() for g in range(n)] # intermediate channels
|
43 |
+
else: # equal weight.numel() per group
|
44 |
+
b = [c2] + [0] * n
|
45 |
+
a = np.eye(n + 1, n, k=-1)
|
46 |
+
a -= np.roll(a, 1, axis=1)
|
47 |
+
a *= np.array(k) ** 2
|
48 |
+
a[0] = 1
|
49 |
+
c_ = np.linalg.lstsq(a, b, rcond=None)[0].round() # solve for equal weight indices, ax = b
|
50 |
+
|
51 |
+
self.m = nn.ModuleList([
|
52 |
+
nn.Conv2d(c1, int(c_), k, s, k // 2, groups=math.gcd(c1, int(c_)), bias=False) for k, c_ in zip(k, c_)])
|
53 |
+
self.bn = nn.BatchNorm2d(c2)
|
54 |
+
self.act = nn.SiLU()
|
55 |
+
|
56 |
+
def forward(self, x):
|
57 |
+
return self.act(self.bn(torch.cat([m(x) for m in self.m], 1)))
|
58 |
+
|
59 |
+
|
60 |
+
class Ensemble(nn.ModuleList):
|
61 |
+
# Ensemble of models
|
62 |
+
def __init__(self):
|
63 |
+
super().__init__()
|
64 |
+
|
65 |
+
def forward(self, x, augment=False, profile=False, visualize=False):
|
66 |
+
y = [module(x, augment, profile, visualize)[0] for module in self]
|
67 |
+
# y = torch.stack(y).max(0)[0] # max ensemble
|
68 |
+
# y = torch.stack(y).mean(0) # mean ensemble
|
69 |
+
y = torch.cat(y, 1) # nms ensemble
|
70 |
+
return y, None # inference, train output
|
71 |
+
|
72 |
+
|
73 |
+
def attempt_load(weights, device=None, inplace=True, fuse=True):
|
74 |
+
# Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a
|
75 |
+
from models.yolo import Detect, Model
|
76 |
+
|
77 |
+
model = Ensemble()
|
78 |
+
for w in weights if isinstance(weights, list) else [weights]:
|
79 |
+
ckpt = torch.load(attempt_download(w), map_location='cpu') # load
|
80 |
+
ckpt = (ckpt.get('ema') or ckpt['model']).to(device).float() # FP32 model
|
81 |
+
|
82 |
+
# Model compatibility updates
|
83 |
+
if not hasattr(ckpt, 'stride'):
|
84 |
+
ckpt.stride = torch.tensor([32.])
|
85 |
+
if hasattr(ckpt, 'names') and isinstance(ckpt.names, (list, tuple)):
|
86 |
+
ckpt.names = dict(enumerate(ckpt.names)) # convert to dict
|
87 |
+
|
88 |
+
model.append(ckpt.fuse().eval() if fuse and hasattr(ckpt, 'fuse') else ckpt.eval()) # model in eval mode
|
89 |
+
|
90 |
+
# Module updates
|
91 |
+
for m in model.modules():
|
92 |
+
t = type(m)
|
93 |
+
if t in (nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU, Detect, Model):
|
94 |
+
m.inplace = inplace
|
95 |
+
if t is Detect and not isinstance(m.anchor_grid, list):
|
96 |
+
delattr(m, 'anchor_grid')
|
97 |
+
setattr(m, 'anchor_grid', [torch.zeros(1)] * m.nl)
|
98 |
+
elif t is nn.Upsample and not hasattr(m, 'recompute_scale_factor'):
|
99 |
+
m.recompute_scale_factor = None # torch 1.11.0 compatibility
|
100 |
+
|
101 |
+
# Return model
|
102 |
+
if len(model) == 1:
|
103 |
+
return model[-1]
|
104 |
+
|
105 |
+
# Return detection ensemble
|
106 |
+
print(f'Ensemble created with {weights}\n')
|
107 |
+
for k in 'names', 'nc', 'yaml':
|
108 |
+
setattr(model, k, getattr(model[0], k))
|
109 |
+
model.stride = model[torch.argmax(torch.tensor([m.stride.max() for m in model])).int()].stride # max stride
|
110 |
+
assert all(model[0].nc == m.nc for m in model), f'Models have different class counts: {[m.nc for m in model]}'
|
111 |
+
return model
|
models/hub/anchors.yaml
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
# Default anchors for COCO data
|
3 |
+
|
4 |
+
|
5 |
+
# P5 -------------------------------------------------------------------------------------------------------------------
|
6 |
+
# P5-640:
|
7 |
+
anchors_p5_640:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
|
13 |
+
# P6 -------------------------------------------------------------------------------------------------------------------
|
14 |
+
# P6-640: thr=0.25: 0.9964 BPR, 5.54 anchors past thr, n=12, img_size=640, metric_all=0.281/0.716-mean/best, past_thr=0.469-mean: 9,11, 21,19, 17,41, 43,32, 39,70, 86,64, 65,131, 134,130, 120,265, 282,180, 247,354, 512,387
|
15 |
+
anchors_p6_640:
|
16 |
+
- [9,11, 21,19, 17,41] # P3/8
|
17 |
+
- [43,32, 39,70, 86,64] # P4/16
|
18 |
+
- [65,131, 134,130, 120,265] # P5/32
|
19 |
+
- [282,180, 247,354, 512,387] # P6/64
|
20 |
+
|
21 |
+
# P6-1280: thr=0.25: 0.9950 BPR, 5.55 anchors past thr, n=12, img_size=1280, metric_all=0.281/0.714-mean/best, past_thr=0.468-mean: 19,27, 44,40, 38,94, 96,68, 86,152, 180,137, 140,301, 303,264, 238,542, 436,615, 739,380, 925,792
|
22 |
+
anchors_p6_1280:
|
23 |
+
- [19,27, 44,40, 38,94] # P3/8
|
24 |
+
- [96,68, 86,152, 180,137] # P4/16
|
25 |
+
- [140,301, 303,264, 238,542] # P5/32
|
26 |
+
- [436,615, 739,380, 925,792] # P6/64
|
27 |
+
|
28 |
+
# P6-1920: thr=0.25: 0.9950 BPR, 5.55 anchors past thr, n=12, img_size=1920, metric_all=0.281/0.714-mean/best, past_thr=0.468-mean: 28,41, 67,59, 57,141, 144,103, 129,227, 270,205, 209,452, 455,396, 358,812, 653,922, 1109,570, 1387,1187
|
29 |
+
anchors_p6_1920:
|
30 |
+
- [28,41, 67,59, 57,141] # P3/8
|
31 |
+
- [144,103, 129,227, 270,205] # P4/16
|
32 |
+
- [209,452, 455,396, 358,812] # P5/32
|
33 |
+
- [653,922, 1109,570, 1387,1187] # P6/64
|
34 |
+
|
35 |
+
|
36 |
+
# P7 -------------------------------------------------------------------------------------------------------------------
|
37 |
+
# P7-640: thr=0.25: 0.9962 BPR, 6.76 anchors past thr, n=15, img_size=640, metric_all=0.275/0.733-mean/best, past_thr=0.466-mean: 11,11, 13,30, 29,20, 30,46, 61,38, 39,92, 78,80, 146,66, 79,163, 149,150, 321,143, 157,303, 257,402, 359,290, 524,372
|
38 |
+
anchors_p7_640:
|
39 |
+
- [11,11, 13,30, 29,20] # P3/8
|
40 |
+
- [30,46, 61,38, 39,92] # P4/16
|
41 |
+
- [78,80, 146,66, 79,163] # P5/32
|
42 |
+
- [149,150, 321,143, 157,303] # P6/64
|
43 |
+
- [257,402, 359,290, 524,372] # P7/128
|
44 |
+
|
45 |
+
# P7-1280: thr=0.25: 0.9968 BPR, 6.71 anchors past thr, n=15, img_size=1280, metric_all=0.273/0.732-mean/best, past_thr=0.463-mean: 19,22, 54,36, 32,77, 70,83, 138,71, 75,173, 165,159, 148,334, 375,151, 334,317, 251,626, 499,474, 750,326, 534,814, 1079,818
|
46 |
+
anchors_p7_1280:
|
47 |
+
- [19,22, 54,36, 32,77] # P3/8
|
48 |
+
- [70,83, 138,71, 75,173] # P4/16
|
49 |
+
- [165,159, 148,334, 375,151] # P5/32
|
50 |
+
- [334,317, 251,626, 499,474] # P6/64
|
51 |
+
- [750,326, 534,814, 1079,818] # P7/128
|
52 |
+
|
53 |
+
# P7-1920: thr=0.25: 0.9968 BPR, 6.71 anchors past thr, n=15, img_size=1920, metric_all=0.273/0.732-mean/best, past_thr=0.463-mean: 29,34, 81,55, 47,115, 105,124, 207,107, 113,259, 247,238, 222,500, 563,227, 501,476, 376,939, 749,711, 1126,489, 801,1222, 1618,1227
|
54 |
+
anchors_p7_1920:
|
55 |
+
- [29,34, 81,55, 47,115] # P3/8
|
56 |
+
- [105,124, 207,107, 113,259] # P4/16
|
57 |
+
- [247,238, 222,500, 563,227] # P5/32
|
58 |
+
- [501,476, 376,939, 749,711] # P6/64
|
59 |
+
- [1126,489, 801,1222, 1618,1227] # P7/128
|
models/hub/yolov3-spp.yaml
ADDED
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# darknet53 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [32, 3, 1]], # 0
|
16 |
+
[-1, 1, Conv, [64, 3, 2]], # 1-P1/2
|
17 |
+
[-1, 1, Bottleneck, [64]],
|
18 |
+
[-1, 1, Conv, [128, 3, 2]], # 3-P2/4
|
19 |
+
[-1, 2, Bottleneck, [128]],
|
20 |
+
[-1, 1, Conv, [256, 3, 2]], # 5-P3/8
|
21 |
+
[-1, 8, Bottleneck, [256]],
|
22 |
+
[-1, 1, Conv, [512, 3, 2]], # 7-P4/16
|
23 |
+
[-1, 8, Bottleneck, [512]],
|
24 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P5/32
|
25 |
+
[-1, 4, Bottleneck, [1024]], # 10
|
26 |
+
]
|
27 |
+
|
28 |
+
# YOLOv3-SPP head
|
29 |
+
head:
|
30 |
+
[[-1, 1, Bottleneck, [1024, False]],
|
31 |
+
[-1, 1, SPP, [512, [5, 9, 13]]],
|
32 |
+
[-1, 1, Conv, [1024, 3, 1]],
|
33 |
+
[-1, 1, Conv, [512, 1, 1]],
|
34 |
+
[-1, 1, Conv, [1024, 3, 1]], # 15 (P5/32-large)
|
35 |
+
|
36 |
+
[-2, 1, Conv, [256, 1, 1]],
|
37 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
38 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P4
|
39 |
+
[-1, 1, Bottleneck, [512, False]],
|
40 |
+
[-1, 1, Bottleneck, [512, False]],
|
41 |
+
[-1, 1, Conv, [256, 1, 1]],
|
42 |
+
[-1, 1, Conv, [512, 3, 1]], # 22 (P4/16-medium)
|
43 |
+
|
44 |
+
[-2, 1, Conv, [128, 1, 1]],
|
45 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
46 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P3
|
47 |
+
[-1, 1, Bottleneck, [256, False]],
|
48 |
+
[-1, 2, Bottleneck, [256, False]], # 27 (P3/8-small)
|
49 |
+
|
50 |
+
[[27, 22, 15], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)
|
51 |
+
]
|
models/hub/yolov3-tiny.yaml
ADDED
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,14, 23,27, 37,58] # P4/16
|
9 |
+
- [81,82, 135,169, 344,319] # P5/32
|
10 |
+
|
11 |
+
# YOLOv3-tiny backbone
|
12 |
+
backbone:
|
13 |
+
# [from, number, module, args]
|
14 |
+
[[-1, 1, Conv, [16, 3, 1]], # 0
|
15 |
+
[-1, 1, nn.MaxPool2d, [2, 2, 0]], # 1-P1/2
|
16 |
+
[-1, 1, Conv, [32, 3, 1]],
|
17 |
+
[-1, 1, nn.MaxPool2d, [2, 2, 0]], # 3-P2/4
|
18 |
+
[-1, 1, Conv, [64, 3, 1]],
|
19 |
+
[-1, 1, nn.MaxPool2d, [2, 2, 0]], # 5-P3/8
|
20 |
+
[-1, 1, Conv, [128, 3, 1]],
|
21 |
+
[-1, 1, nn.MaxPool2d, [2, 2, 0]], # 7-P4/16
|
22 |
+
[-1, 1, Conv, [256, 3, 1]],
|
23 |
+
[-1, 1, nn.MaxPool2d, [2, 2, 0]], # 9-P5/32
|
24 |
+
[-1, 1, Conv, [512, 3, 1]],
|
25 |
+
[-1, 1, nn.ZeroPad2d, [[0, 1, 0, 1]]], # 11
|
26 |
+
[-1, 1, nn.MaxPool2d, [2, 1, 0]], # 12
|
27 |
+
]
|
28 |
+
|
29 |
+
# YOLOv3-tiny head
|
30 |
+
head:
|
31 |
+
[[-1, 1, Conv, [1024, 3, 1]],
|
32 |
+
[-1, 1, Conv, [256, 1, 1]],
|
33 |
+
[-1, 1, Conv, [512, 3, 1]], # 15 (P5/32-large)
|
34 |
+
|
35 |
+
[-2, 1, Conv, [128, 1, 1]],
|
36 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
37 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P4
|
38 |
+
[-1, 1, Conv, [256, 3, 1]], # 19 (P4/16-medium)
|
39 |
+
|
40 |
+
[[19, 15], 1, Detect, [nc, anchors]], # Detect(P4, P5)
|
41 |
+
]
|
models/hub/yolov3.yaml
ADDED
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# darknet53 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [32, 3, 1]], # 0
|
16 |
+
[-1, 1, Conv, [64, 3, 2]], # 1-P1/2
|
17 |
+
[-1, 1, Bottleneck, [64]],
|
18 |
+
[-1, 1, Conv, [128, 3, 2]], # 3-P2/4
|
19 |
+
[-1, 2, Bottleneck, [128]],
|
20 |
+
[-1, 1, Conv, [256, 3, 2]], # 5-P3/8
|
21 |
+
[-1, 8, Bottleneck, [256]],
|
22 |
+
[-1, 1, Conv, [512, 3, 2]], # 7-P4/16
|
23 |
+
[-1, 8, Bottleneck, [512]],
|
24 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P5/32
|
25 |
+
[-1, 4, Bottleneck, [1024]], # 10
|
26 |
+
]
|
27 |
+
|
28 |
+
# YOLOv3 head
|
29 |
+
head:
|
30 |
+
[[-1, 1, Bottleneck, [1024, False]],
|
31 |
+
[-1, 1, Conv, [512, 1, 1]],
|
32 |
+
[-1, 1, Conv, [1024, 3, 1]],
|
33 |
+
[-1, 1, Conv, [512, 1, 1]],
|
34 |
+
[-1, 1, Conv, [1024, 3, 1]], # 15 (P5/32-large)
|
35 |
+
|
36 |
+
[-2, 1, Conv, [256, 1, 1]],
|
37 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
38 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P4
|
39 |
+
[-1, 1, Bottleneck, [512, False]],
|
40 |
+
[-1, 1, Bottleneck, [512, False]],
|
41 |
+
[-1, 1, Conv, [256, 1, 1]],
|
42 |
+
[-1, 1, Conv, [512, 3, 1]], # 22 (P4/16-medium)
|
43 |
+
|
44 |
+
[-2, 1, Conv, [128, 1, 1]],
|
45 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
46 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P3
|
47 |
+
[-1, 1, Bottleneck, [256, False]],
|
48 |
+
[-1, 2, Bottleneck, [256, False]], # 27 (P3/8-small)
|
49 |
+
|
50 |
+
[[27, 22, 15], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)
|
51 |
+
]
|
models/hub/yolov5-bifpn.yaml
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# YOLOv5 v6.0 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
16 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
17 |
+
[-1, 3, C3, [128]],
|
18 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
19 |
+
[-1, 6, C3, [256]],
|
20 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
21 |
+
[-1, 9, C3, [512]],
|
22 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
23 |
+
[-1, 3, C3, [1024]],
|
24 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
25 |
+
]
|
26 |
+
|
27 |
+
# YOLOv5 v6.0 BiFPN head
|
28 |
+
head:
|
29 |
+
[[-1, 1, Conv, [512, 1, 1]],
|
30 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
31 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
32 |
+
[-1, 3, C3, [512, False]], # 13
|
33 |
+
|
34 |
+
[-1, 1, Conv, [256, 1, 1]],
|
35 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
36 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
37 |
+
[-1, 3, C3, [256, False]], # 17 (P3/8-small)
|
38 |
+
|
39 |
+
[-1, 1, Conv, [256, 3, 2]],
|
40 |
+
[[-1, 14, 6], 1, Concat, [1]], # cat P4 <--- BiFPN change
|
41 |
+
[-1, 3, C3, [512, False]], # 20 (P4/16-medium)
|
42 |
+
|
43 |
+
[-1, 1, Conv, [512, 3, 2]],
|
44 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
45 |
+
[-1, 3, C3, [1024, False]], # 23 (P5/32-large)
|
46 |
+
|
47 |
+
[[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)
|
48 |
+
]
|
models/hub/yolov5-fpn.yaml
ADDED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# YOLOv5 v6.0 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
16 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
17 |
+
[-1, 3, C3, [128]],
|
18 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
19 |
+
[-1, 6, C3, [256]],
|
20 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
21 |
+
[-1, 9, C3, [512]],
|
22 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
23 |
+
[-1, 3, C3, [1024]],
|
24 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
25 |
+
]
|
26 |
+
|
27 |
+
# YOLOv5 v6.0 FPN head
|
28 |
+
head:
|
29 |
+
[[-1, 3, C3, [1024, False]], # 10 (P5/32-large)
|
30 |
+
|
31 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
32 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
33 |
+
[-1, 1, Conv, [512, 1, 1]],
|
34 |
+
[-1, 3, C3, [512, False]], # 14 (P4/16-medium)
|
35 |
+
|
36 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
37 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
38 |
+
[-1, 1, Conv, [256, 1, 1]],
|
39 |
+
[-1, 3, C3, [256, False]], # 18 (P3/8-small)
|
40 |
+
|
41 |
+
[[18, 14, 10], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)
|
42 |
+
]
|
models/hub/yolov5-p2.yaml
ADDED
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors: 3 # AutoAnchor evolves 3 anchors per P output layer
|
8 |
+
|
9 |
+
# YOLOv5 v6.0 backbone
|
10 |
+
backbone:
|
11 |
+
# [from, number, module, args]
|
12 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
13 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
14 |
+
[-1, 3, C3, [128]],
|
15 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
16 |
+
[-1, 6, C3, [256]],
|
17 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
18 |
+
[-1, 9, C3, [512]],
|
19 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
20 |
+
[-1, 3, C3, [1024]],
|
21 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
22 |
+
]
|
23 |
+
|
24 |
+
# YOLOv5 v6.0 head with (P2, P3, P4, P5) outputs
|
25 |
+
head:
|
26 |
+
[[-1, 1, Conv, [512, 1, 1]],
|
27 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
28 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
29 |
+
[-1, 3, C3, [512, False]], # 13
|
30 |
+
|
31 |
+
[-1, 1, Conv, [256, 1, 1]],
|
32 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
33 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
34 |
+
[-1, 3, C3, [256, False]], # 17 (P3/8-small)
|
35 |
+
|
36 |
+
[-1, 1, Conv, [128, 1, 1]],
|
37 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
38 |
+
[[-1, 2], 1, Concat, [1]], # cat backbone P2
|
39 |
+
[-1, 1, C3, [128, False]], # 21 (P2/4-xsmall)
|
40 |
+
|
41 |
+
[-1, 1, Conv, [128, 3, 2]],
|
42 |
+
[[-1, 18], 1, Concat, [1]], # cat head P3
|
43 |
+
[-1, 3, C3, [256, False]], # 24 (P3/8-small)
|
44 |
+
|
45 |
+
[-1, 1, Conv, [256, 3, 2]],
|
46 |
+
[[-1, 14], 1, Concat, [1]], # cat head P4
|
47 |
+
[-1, 3, C3, [512, False]], # 27 (P4/16-medium)
|
48 |
+
|
49 |
+
[-1, 1, Conv, [512, 3, 2]],
|
50 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
51 |
+
[-1, 3, C3, [1024, False]], # 30 (P5/32-large)
|
52 |
+
|
53 |
+
[[21, 24, 27, 30], 1, Detect, [nc, anchors]], # Detect(P2, P3, P4, P5)
|
54 |
+
]
|
models/hub/yolov5-p34.yaml
ADDED
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 0.33 # model depth multiple
|
6 |
+
width_multiple: 0.50 # layer channel multiple
|
7 |
+
anchors: 3 # AutoAnchor evolves 3 anchors per P output layer
|
8 |
+
|
9 |
+
# YOLOv5 v6.0 backbone
|
10 |
+
backbone:
|
11 |
+
# [from, number, module, args]
|
12 |
+
[ [ -1, 1, Conv, [ 64, 6, 2, 2 ] ], # 0-P1/2
|
13 |
+
[ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4
|
14 |
+
[ -1, 3, C3, [ 128 ] ],
|
15 |
+
[ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8
|
16 |
+
[ -1, 6, C3, [ 256 ] ],
|
17 |
+
[ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16
|
18 |
+
[ -1, 9, C3, [ 512 ] ],
|
19 |
+
[ -1, 1, Conv, [ 1024, 3, 2 ] ], # 7-P5/32
|
20 |
+
[ -1, 3, C3, [ 1024 ] ],
|
21 |
+
[ -1, 1, SPPF, [ 1024, 5 ] ], # 9
|
22 |
+
]
|
23 |
+
|
24 |
+
# YOLOv5 v6.0 head with (P3, P4) outputs
|
25 |
+
head:
|
26 |
+
[ [ -1, 1, Conv, [ 512, 1, 1 ] ],
|
27 |
+
[ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
|
28 |
+
[ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4
|
29 |
+
[ -1, 3, C3, [ 512, False ] ], # 13
|
30 |
+
|
31 |
+
[ -1, 1, Conv, [ 256, 1, 1 ] ],
|
32 |
+
[ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
|
33 |
+
[ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3
|
34 |
+
[ -1, 3, C3, [ 256, False ] ], # 17 (P3/8-small)
|
35 |
+
|
36 |
+
[ -1, 1, Conv, [ 256, 3, 2 ] ],
|
37 |
+
[ [ -1, 14 ], 1, Concat, [ 1 ] ], # cat head P4
|
38 |
+
[ -1, 3, C3, [ 512, False ] ], # 20 (P4/16-medium)
|
39 |
+
|
40 |
+
[ [ 17, 20 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4)
|
41 |
+
]
|
models/hub/yolov5-p6.yaml
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors: 3 # AutoAnchor evolves 3 anchors per P output layer
|
8 |
+
|
9 |
+
# YOLOv5 v6.0 backbone
|
10 |
+
backbone:
|
11 |
+
# [from, number, module, args]
|
12 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
13 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
14 |
+
[-1, 3, C3, [128]],
|
15 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
16 |
+
[-1, 6, C3, [256]],
|
17 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
18 |
+
[-1, 9, C3, [512]],
|
19 |
+
[-1, 1, Conv, [768, 3, 2]], # 7-P5/32
|
20 |
+
[-1, 3, C3, [768]],
|
21 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P6/64
|
22 |
+
[-1, 3, C3, [1024]],
|
23 |
+
[-1, 1, SPPF, [1024, 5]], # 11
|
24 |
+
]
|
25 |
+
|
26 |
+
# YOLOv5 v6.0 head with (P3, P4, P5, P6) outputs
|
27 |
+
head:
|
28 |
+
[[-1, 1, Conv, [768, 1, 1]],
|
29 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
30 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P5
|
31 |
+
[-1, 3, C3, [768, False]], # 15
|
32 |
+
|
33 |
+
[-1, 1, Conv, [512, 1, 1]],
|
34 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
35 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
36 |
+
[-1, 3, C3, [512, False]], # 19
|
37 |
+
|
38 |
+
[-1, 1, Conv, [256, 1, 1]],
|
39 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
40 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
41 |
+
[-1, 3, C3, [256, False]], # 23 (P3/8-small)
|
42 |
+
|
43 |
+
[-1, 1, Conv, [256, 3, 2]],
|
44 |
+
[[-1, 20], 1, Concat, [1]], # cat head P4
|
45 |
+
[-1, 3, C3, [512, False]], # 26 (P4/16-medium)
|
46 |
+
|
47 |
+
[-1, 1, Conv, [512, 3, 2]],
|
48 |
+
[[-1, 16], 1, Concat, [1]], # cat head P5
|
49 |
+
[-1, 3, C3, [768, False]], # 29 (P5/32-large)
|
50 |
+
|
51 |
+
[-1, 1, Conv, [768, 3, 2]],
|
52 |
+
[[-1, 12], 1, Concat, [1]], # cat head P6
|
53 |
+
[-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge)
|
54 |
+
|
55 |
+
[[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6)
|
56 |
+
]
|
models/hub/yolov5-p7.yaml
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors: 3 # AutoAnchor evolves 3 anchors per P output layer
|
8 |
+
|
9 |
+
# YOLOv5 v6.0 backbone
|
10 |
+
backbone:
|
11 |
+
# [from, number, module, args]
|
12 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
13 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
14 |
+
[-1, 3, C3, [128]],
|
15 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
16 |
+
[-1, 6, C3, [256]],
|
17 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
18 |
+
[-1, 9, C3, [512]],
|
19 |
+
[-1, 1, Conv, [768, 3, 2]], # 7-P5/32
|
20 |
+
[-1, 3, C3, [768]],
|
21 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P6/64
|
22 |
+
[-1, 3, C3, [1024]],
|
23 |
+
[-1, 1, Conv, [1280, 3, 2]], # 11-P7/128
|
24 |
+
[-1, 3, C3, [1280]],
|
25 |
+
[-1, 1, SPPF, [1280, 5]], # 13
|
26 |
+
]
|
27 |
+
|
28 |
+
# YOLOv5 v6.0 head with (P3, P4, P5, P6, P7) outputs
|
29 |
+
head:
|
30 |
+
[[-1, 1, Conv, [1024, 1, 1]],
|
31 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
32 |
+
[[-1, 10], 1, Concat, [1]], # cat backbone P6
|
33 |
+
[-1, 3, C3, [1024, False]], # 17
|
34 |
+
|
35 |
+
[-1, 1, Conv, [768, 1, 1]],
|
36 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
37 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P5
|
38 |
+
[-1, 3, C3, [768, False]], # 21
|
39 |
+
|
40 |
+
[-1, 1, Conv, [512, 1, 1]],
|
41 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
42 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
43 |
+
[-1, 3, C3, [512, False]], # 25
|
44 |
+
|
45 |
+
[-1, 1, Conv, [256, 1, 1]],
|
46 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
47 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
48 |
+
[-1, 3, C3, [256, False]], # 29 (P3/8-small)
|
49 |
+
|
50 |
+
[-1, 1, Conv, [256, 3, 2]],
|
51 |
+
[[-1, 26], 1, Concat, [1]], # cat head P4
|
52 |
+
[-1, 3, C3, [512, False]], # 32 (P4/16-medium)
|
53 |
+
|
54 |
+
[-1, 1, Conv, [512, 3, 2]],
|
55 |
+
[[-1, 22], 1, Concat, [1]], # cat head P5
|
56 |
+
[-1, 3, C3, [768, False]], # 35 (P5/32-large)
|
57 |
+
|
58 |
+
[-1, 1, Conv, [768, 3, 2]],
|
59 |
+
[[-1, 18], 1, Concat, [1]], # cat head P6
|
60 |
+
[-1, 3, C3, [1024, False]], # 38 (P6/64-xlarge)
|
61 |
+
|
62 |
+
[-1, 1, Conv, [1024, 3, 2]],
|
63 |
+
[[-1, 14], 1, Concat, [1]], # cat head P7
|
64 |
+
[-1, 3, C3, [1280, False]], # 41 (P7/128-xxlarge)
|
65 |
+
|
66 |
+
[[29, 32, 35, 38, 41], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6, P7)
|
67 |
+
]
|
models/hub/yolov5-panet.yaml
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# YOLOv5 v6.0 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
16 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
17 |
+
[-1, 3, C3, [128]],
|
18 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
19 |
+
[-1, 6, C3, [256]],
|
20 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
21 |
+
[-1, 9, C3, [512]],
|
22 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
23 |
+
[-1, 3, C3, [1024]],
|
24 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
25 |
+
]
|
26 |
+
|
27 |
+
# YOLOv5 v6.0 PANet head
|
28 |
+
head:
|
29 |
+
[[-1, 1, Conv, [512, 1, 1]],
|
30 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
31 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
32 |
+
[-1, 3, C3, [512, False]], # 13
|
33 |
+
|
34 |
+
[-1, 1, Conv, [256, 1, 1]],
|
35 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
36 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
37 |
+
[-1, 3, C3, [256, False]], # 17 (P3/8-small)
|
38 |
+
|
39 |
+
[-1, 1, Conv, [256, 3, 2]],
|
40 |
+
[[-1, 14], 1, Concat, [1]], # cat head P4
|
41 |
+
[-1, 3, C3, [512, False]], # 20 (P4/16-medium)
|
42 |
+
|
43 |
+
[-1, 1, Conv, [512, 3, 2]],
|
44 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
45 |
+
[-1, 3, C3, [1024, False]], # 23 (P5/32-large)
|
46 |
+
|
47 |
+
[[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)
|
48 |
+
]
|
models/hub/yolov5l6.yaml
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [19,27, 44,40, 38,94] # P3/8
|
9 |
+
- [96,68, 86,152, 180,137] # P4/16
|
10 |
+
- [140,301, 303,264, 238,542] # P5/32
|
11 |
+
- [436,615, 739,380, 925,792] # P6/64
|
12 |
+
|
13 |
+
# YOLOv5 v6.0 backbone
|
14 |
+
backbone:
|
15 |
+
# [from, number, module, args]
|
16 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
17 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
18 |
+
[-1, 3, C3, [128]],
|
19 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
20 |
+
[-1, 6, C3, [256]],
|
21 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
22 |
+
[-1, 9, C3, [512]],
|
23 |
+
[-1, 1, Conv, [768, 3, 2]], # 7-P5/32
|
24 |
+
[-1, 3, C3, [768]],
|
25 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P6/64
|
26 |
+
[-1, 3, C3, [1024]],
|
27 |
+
[-1, 1, SPPF, [1024, 5]], # 11
|
28 |
+
]
|
29 |
+
|
30 |
+
# YOLOv5 v6.0 head
|
31 |
+
head:
|
32 |
+
[[-1, 1, Conv, [768, 1, 1]],
|
33 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
34 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P5
|
35 |
+
[-1, 3, C3, [768, False]], # 15
|
36 |
+
|
37 |
+
[-1, 1, Conv, [512, 1, 1]],
|
38 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
39 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
40 |
+
[-1, 3, C3, [512, False]], # 19
|
41 |
+
|
42 |
+
[-1, 1, Conv, [256, 1, 1]],
|
43 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
44 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
45 |
+
[-1, 3, C3, [256, False]], # 23 (P3/8-small)
|
46 |
+
|
47 |
+
[-1, 1, Conv, [256, 3, 2]],
|
48 |
+
[[-1, 20], 1, Concat, [1]], # cat head P4
|
49 |
+
[-1, 3, C3, [512, False]], # 26 (P4/16-medium)
|
50 |
+
|
51 |
+
[-1, 1, Conv, [512, 3, 2]],
|
52 |
+
[[-1, 16], 1, Concat, [1]], # cat head P5
|
53 |
+
[-1, 3, C3, [768, False]], # 29 (P5/32-large)
|
54 |
+
|
55 |
+
[-1, 1, Conv, [768, 3, 2]],
|
56 |
+
[[-1, 12], 1, Concat, [1]], # cat head P6
|
57 |
+
[-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge)
|
58 |
+
|
59 |
+
[[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6)
|
60 |
+
]
|
models/hub/yolov5m6.yaml
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 0.67 # model depth multiple
|
6 |
+
width_multiple: 0.75 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [19,27, 44,40, 38,94] # P3/8
|
9 |
+
- [96,68, 86,152, 180,137] # P4/16
|
10 |
+
- [140,301, 303,264, 238,542] # P5/32
|
11 |
+
- [436,615, 739,380, 925,792] # P6/64
|
12 |
+
|
13 |
+
# YOLOv5 v6.0 backbone
|
14 |
+
backbone:
|
15 |
+
# [from, number, module, args]
|
16 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
17 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
18 |
+
[-1, 3, C3, [128]],
|
19 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
20 |
+
[-1, 6, C3, [256]],
|
21 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
22 |
+
[-1, 9, C3, [512]],
|
23 |
+
[-1, 1, Conv, [768, 3, 2]], # 7-P5/32
|
24 |
+
[-1, 3, C3, [768]],
|
25 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P6/64
|
26 |
+
[-1, 3, C3, [1024]],
|
27 |
+
[-1, 1, SPPF, [1024, 5]], # 11
|
28 |
+
]
|
29 |
+
|
30 |
+
# YOLOv5 v6.0 head
|
31 |
+
head:
|
32 |
+
[[-1, 1, Conv, [768, 1, 1]],
|
33 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
34 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P5
|
35 |
+
[-1, 3, C3, [768, False]], # 15
|
36 |
+
|
37 |
+
[-1, 1, Conv, [512, 1, 1]],
|
38 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
39 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
40 |
+
[-1, 3, C3, [512, False]], # 19
|
41 |
+
|
42 |
+
[-1, 1, Conv, [256, 1, 1]],
|
43 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
44 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
45 |
+
[-1, 3, C3, [256, False]], # 23 (P3/8-small)
|
46 |
+
|
47 |
+
[-1, 1, Conv, [256, 3, 2]],
|
48 |
+
[[-1, 20], 1, Concat, [1]], # cat head P4
|
49 |
+
[-1, 3, C3, [512, False]], # 26 (P4/16-medium)
|
50 |
+
|
51 |
+
[-1, 1, Conv, [512, 3, 2]],
|
52 |
+
[[-1, 16], 1, Concat, [1]], # cat head P5
|
53 |
+
[-1, 3, C3, [768, False]], # 29 (P5/32-large)
|
54 |
+
|
55 |
+
[-1, 1, Conv, [768, 3, 2]],
|
56 |
+
[[-1, 12], 1, Concat, [1]], # cat head P6
|
57 |
+
[-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge)
|
58 |
+
|
59 |
+
[[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6)
|
60 |
+
]
|
models/hub/yolov5n6.yaml
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 0.33 # model depth multiple
|
6 |
+
width_multiple: 0.25 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [19,27, 44,40, 38,94] # P3/8
|
9 |
+
- [96,68, 86,152, 180,137] # P4/16
|
10 |
+
- [140,301, 303,264, 238,542] # P5/32
|
11 |
+
- [436,615, 739,380, 925,792] # P6/64
|
12 |
+
|
13 |
+
# YOLOv5 v6.0 backbone
|
14 |
+
backbone:
|
15 |
+
# [from, number, module, args]
|
16 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
17 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
18 |
+
[-1, 3, C3, [128]],
|
19 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
20 |
+
[-1, 6, C3, [256]],
|
21 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
22 |
+
[-1, 9, C3, [512]],
|
23 |
+
[-1, 1, Conv, [768, 3, 2]], # 7-P5/32
|
24 |
+
[-1, 3, C3, [768]],
|
25 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P6/64
|
26 |
+
[-1, 3, C3, [1024]],
|
27 |
+
[-1, 1, SPPF, [1024, 5]], # 11
|
28 |
+
]
|
29 |
+
|
30 |
+
# YOLOv5 v6.0 head
|
31 |
+
head:
|
32 |
+
[[-1, 1, Conv, [768, 1, 1]],
|
33 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
34 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P5
|
35 |
+
[-1, 3, C3, [768, False]], # 15
|
36 |
+
|
37 |
+
[-1, 1, Conv, [512, 1, 1]],
|
38 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
39 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
40 |
+
[-1, 3, C3, [512, False]], # 19
|
41 |
+
|
42 |
+
[-1, 1, Conv, [256, 1, 1]],
|
43 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
44 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
45 |
+
[-1, 3, C3, [256, False]], # 23 (P3/8-small)
|
46 |
+
|
47 |
+
[-1, 1, Conv, [256, 3, 2]],
|
48 |
+
[[-1, 20], 1, Concat, [1]], # cat head P4
|
49 |
+
[-1, 3, C3, [512, False]], # 26 (P4/16-medium)
|
50 |
+
|
51 |
+
[-1, 1, Conv, [512, 3, 2]],
|
52 |
+
[[-1, 16], 1, Concat, [1]], # cat head P5
|
53 |
+
[-1, 3, C3, [768, False]], # 29 (P5/32-large)
|
54 |
+
|
55 |
+
[-1, 1, Conv, [768, 3, 2]],
|
56 |
+
[[-1, 12], 1, Concat, [1]], # cat head P6
|
57 |
+
[-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge)
|
58 |
+
|
59 |
+
[[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6)
|
60 |
+
]
|
models/hub/yolov5s-LeakyReLU.yaml
ADDED
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
activation: nn.LeakyReLU(0.1) # <----- Conv() activation used throughout entire YOLOv5 model
|
6 |
+
depth_multiple: 0.33 # model depth multiple
|
7 |
+
width_multiple: 0.50 # layer channel multiple
|
8 |
+
anchors:
|
9 |
+
- [10,13, 16,30, 33,23] # P3/8
|
10 |
+
- [30,61, 62,45, 59,119] # P4/16
|
11 |
+
- [116,90, 156,198, 373,326] # P5/32
|
12 |
+
|
13 |
+
# YOLOv5 v6.0 backbone
|
14 |
+
backbone:
|
15 |
+
# [from, number, module, args]
|
16 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
17 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
18 |
+
[-1, 3, C3, [128]],
|
19 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
20 |
+
[-1, 6, C3, [256]],
|
21 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
22 |
+
[-1, 9, C3, [512]],
|
23 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
24 |
+
[-1, 3, C3, [1024]],
|
25 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
26 |
+
]
|
27 |
+
|
28 |
+
# YOLOv5 v6.0 head
|
29 |
+
head:
|
30 |
+
[[-1, 1, Conv, [512, 1, 1]],
|
31 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
32 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
33 |
+
[-1, 3, C3, [512, False]], # 13
|
34 |
+
|
35 |
+
[-1, 1, Conv, [256, 1, 1]],
|
36 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
37 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
38 |
+
[-1, 3, C3, [256, False]], # 17 (P3/8-small)
|
39 |
+
|
40 |
+
[-1, 1, Conv, [256, 3, 2]],
|
41 |
+
[[-1, 14], 1, Concat, [1]], # cat head P4
|
42 |
+
[-1, 3, C3, [512, False]], # 20 (P4/16-medium)
|
43 |
+
|
44 |
+
[-1, 1, Conv, [512, 3, 2]],
|
45 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
46 |
+
[-1, 3, C3, [1024, False]], # 23 (P5/32-large)
|
47 |
+
|
48 |
+
[[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)
|
49 |
+
]
|
models/hub/yolov5s-ghost.yaml
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 0.33 # model depth multiple
|
6 |
+
width_multiple: 0.50 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# YOLOv5 v6.0 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
16 |
+
[-1, 1, GhostConv, [128, 3, 2]], # 1-P2/4
|
17 |
+
[-1, 3, C3Ghost, [128]],
|
18 |
+
[-1, 1, GhostConv, [256, 3, 2]], # 3-P3/8
|
19 |
+
[-1, 6, C3Ghost, [256]],
|
20 |
+
[-1, 1, GhostConv, [512, 3, 2]], # 5-P4/16
|
21 |
+
[-1, 9, C3Ghost, [512]],
|
22 |
+
[-1, 1, GhostConv, [1024, 3, 2]], # 7-P5/32
|
23 |
+
[-1, 3, C3Ghost, [1024]],
|
24 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
25 |
+
]
|
26 |
+
|
27 |
+
# YOLOv5 v6.0 head
|
28 |
+
head:
|
29 |
+
[[-1, 1, GhostConv, [512, 1, 1]],
|
30 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
31 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
32 |
+
[-1, 3, C3Ghost, [512, False]], # 13
|
33 |
+
|
34 |
+
[-1, 1, GhostConv, [256, 1, 1]],
|
35 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
36 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
37 |
+
[-1, 3, C3Ghost, [256, False]], # 17 (P3/8-small)
|
38 |
+
|
39 |
+
[-1, 1, GhostConv, [256, 3, 2]],
|
40 |
+
[[-1, 14], 1, Concat, [1]], # cat head P4
|
41 |
+
[-1, 3, C3Ghost, [512, False]], # 20 (P4/16-medium)
|
42 |
+
|
43 |
+
[-1, 1, GhostConv, [512, 3, 2]],
|
44 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
45 |
+
[-1, 3, C3Ghost, [1024, False]], # 23 (P5/32-large)
|
46 |
+
|
47 |
+
[[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)
|
48 |
+
]
|
models/hub/yolov5s-transformer.yaml
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 0.33 # model depth multiple
|
6 |
+
width_multiple: 0.50 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# YOLOv5 v6.0 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
16 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
17 |
+
[-1, 3, C3, [128]],
|
18 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
19 |
+
[-1, 6, C3, [256]],
|
20 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
21 |
+
[-1, 9, C3, [512]],
|
22 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
23 |
+
[-1, 3, C3TR, [1024]], # 9 <--- C3TR() Transformer module
|
24 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
25 |
+
]
|
26 |
+
|
27 |
+
# YOLOv5 v6.0 head
|
28 |
+
head:
|
29 |
+
[[-1, 1, Conv, [512, 1, 1]],
|
30 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
31 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
32 |
+
[-1, 3, C3, [512, False]], # 13
|
33 |
+
|
34 |
+
[-1, 1, Conv, [256, 1, 1]],
|
35 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
36 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
37 |
+
[-1, 3, C3, [256, False]], # 17 (P3/8-small)
|
38 |
+
|
39 |
+
[-1, 1, Conv, [256, 3, 2]],
|
40 |
+
[[-1, 14], 1, Concat, [1]], # cat head P4
|
41 |
+
[-1, 3, C3, [512, False]], # 20 (P4/16-medium)
|
42 |
+
|
43 |
+
[-1, 1, Conv, [512, 3, 2]],
|
44 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
45 |
+
[-1, 3, C3, [1024, False]], # 23 (P5/32-large)
|
46 |
+
|
47 |
+
[[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)
|
48 |
+
]
|
models/hub/yolov5s6.yaml
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 0.33 # model depth multiple
|
6 |
+
width_multiple: 0.50 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [19,27, 44,40, 38,94] # P3/8
|
9 |
+
- [96,68, 86,152, 180,137] # P4/16
|
10 |
+
- [140,301, 303,264, 238,542] # P5/32
|
11 |
+
- [436,615, 739,380, 925,792] # P6/64
|
12 |
+
|
13 |
+
# YOLOv5 v6.0 backbone
|
14 |
+
backbone:
|
15 |
+
# [from, number, module, args]
|
16 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
17 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
18 |
+
[-1, 3, C3, [128]],
|
19 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
20 |
+
[-1, 6, C3, [256]],
|
21 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
22 |
+
[-1, 9, C3, [512]],
|
23 |
+
[-1, 1, Conv, [768, 3, 2]], # 7-P5/32
|
24 |
+
[-1, 3, C3, [768]],
|
25 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P6/64
|
26 |
+
[-1, 3, C3, [1024]],
|
27 |
+
[-1, 1, SPPF, [1024, 5]], # 11
|
28 |
+
]
|
29 |
+
|
30 |
+
# YOLOv5 v6.0 head
|
31 |
+
head:
|
32 |
+
[[-1, 1, Conv, [768, 1, 1]],
|
33 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
34 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P5
|
35 |
+
[-1, 3, C3, [768, False]], # 15
|
36 |
+
|
37 |
+
[-1, 1, Conv, [512, 1, 1]],
|
38 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
39 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
40 |
+
[-1, 3, C3, [512, False]], # 19
|
41 |
+
|
42 |
+
[-1, 1, Conv, [256, 1, 1]],
|
43 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
44 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
45 |
+
[-1, 3, C3, [256, False]], # 23 (P3/8-small)
|
46 |
+
|
47 |
+
[-1, 1, Conv, [256, 3, 2]],
|
48 |
+
[[-1, 20], 1, Concat, [1]], # cat head P4
|
49 |
+
[-1, 3, C3, [512, False]], # 26 (P4/16-medium)
|
50 |
+
|
51 |
+
[-1, 1, Conv, [512, 3, 2]],
|
52 |
+
[[-1, 16], 1, Concat, [1]], # cat head P5
|
53 |
+
[-1, 3, C3, [768, False]], # 29 (P5/32-large)
|
54 |
+
|
55 |
+
[-1, 1, Conv, [768, 3, 2]],
|
56 |
+
[[-1, 12], 1, Concat, [1]], # cat head P6
|
57 |
+
[-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge)
|
58 |
+
|
59 |
+
[[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6)
|
60 |
+
]
|
models/hub/yolov5x6.yaml
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.33 # model depth multiple
|
6 |
+
width_multiple: 1.25 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [19,27, 44,40, 38,94] # P3/8
|
9 |
+
- [96,68, 86,152, 180,137] # P4/16
|
10 |
+
- [140,301, 303,264, 238,542] # P5/32
|
11 |
+
- [436,615, 739,380, 925,792] # P6/64
|
12 |
+
|
13 |
+
# YOLOv5 v6.0 backbone
|
14 |
+
backbone:
|
15 |
+
# [from, number, module, args]
|
16 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
17 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
18 |
+
[-1, 3, C3, [128]],
|
19 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
20 |
+
[-1, 6, C3, [256]],
|
21 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
22 |
+
[-1, 9, C3, [512]],
|
23 |
+
[-1, 1, Conv, [768, 3, 2]], # 7-P5/32
|
24 |
+
[-1, 3, C3, [768]],
|
25 |
+
[-1, 1, Conv, [1024, 3, 2]], # 9-P6/64
|
26 |
+
[-1, 3, C3, [1024]],
|
27 |
+
[-1, 1, SPPF, [1024, 5]], # 11
|
28 |
+
]
|
29 |
+
|
30 |
+
# YOLOv5 v6.0 head
|
31 |
+
head:
|
32 |
+
[[-1, 1, Conv, [768, 1, 1]],
|
33 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
34 |
+
[[-1, 8], 1, Concat, [1]], # cat backbone P5
|
35 |
+
[-1, 3, C3, [768, False]], # 15
|
36 |
+
|
37 |
+
[-1, 1, Conv, [512, 1, 1]],
|
38 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
39 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
40 |
+
[-1, 3, C3, [512, False]], # 19
|
41 |
+
|
42 |
+
[-1, 1, Conv, [256, 1, 1]],
|
43 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
44 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
45 |
+
[-1, 3, C3, [256, False]], # 23 (P3/8-small)
|
46 |
+
|
47 |
+
[-1, 1, Conv, [256, 3, 2]],
|
48 |
+
[[-1, 20], 1, Concat, [1]], # cat head P4
|
49 |
+
[-1, 3, C3, [512, False]], # 26 (P4/16-medium)
|
50 |
+
|
51 |
+
[-1, 1, Conv, [512, 3, 2]],
|
52 |
+
[[-1, 16], 1, Concat, [1]], # cat head P5
|
53 |
+
[-1, 3, C3, [768, False]], # 29 (P5/32-large)
|
54 |
+
|
55 |
+
[-1, 1, Conv, [768, 3, 2]],
|
56 |
+
[[-1, 12], 1, Concat, [1]], # cat head P6
|
57 |
+
[-1, 3, C3, [1024, False]], # 32 (P6/64-xlarge)
|
58 |
+
|
59 |
+
[[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6)
|
60 |
+
]
|
models/segment/yolov5l-seg.yaml
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 1.0 # model depth multiple
|
6 |
+
width_multiple: 1.0 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# YOLOv5 v6.0 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
16 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
17 |
+
[-1, 3, C3, [128]],
|
18 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
19 |
+
[-1, 6, C3, [256]],
|
20 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
21 |
+
[-1, 9, C3, [512]],
|
22 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
23 |
+
[-1, 3, C3, [1024]],
|
24 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
25 |
+
]
|
26 |
+
|
27 |
+
# YOLOv5 v6.0 head
|
28 |
+
head:
|
29 |
+
[[-1, 1, Conv, [512, 1, 1]],
|
30 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
31 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
32 |
+
[-1, 3, C3, [512, False]], # 13
|
33 |
+
|
34 |
+
[-1, 1, Conv, [256, 1, 1]],
|
35 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
36 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
37 |
+
[-1, 3, C3, [256, False]], # 17 (P3/8-small)
|
38 |
+
|
39 |
+
[-1, 1, Conv, [256, 3, 2]],
|
40 |
+
[[-1, 14], 1, Concat, [1]], # cat head P4
|
41 |
+
[-1, 3, C3, [512, False]], # 20 (P4/16-medium)
|
42 |
+
|
43 |
+
[-1, 1, Conv, [512, 3, 2]],
|
44 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
45 |
+
[-1, 3, C3, [1024, False]], # 23 (P5/32-large)
|
46 |
+
|
47 |
+
[[17, 20, 23], 1, Segment, [nc, anchors, 32, 256]], # Detect(P3, P4, P5)
|
48 |
+
]
|
models/segment/yolov5m-seg.yaml
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 0.67 # model depth multiple
|
6 |
+
width_multiple: 0.75 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# YOLOv5 v6.0 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
16 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
17 |
+
[-1, 3, C3, [128]],
|
18 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
19 |
+
[-1, 6, C3, [256]],
|
20 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
21 |
+
[-1, 9, C3, [512]],
|
22 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
23 |
+
[-1, 3, C3, [1024]],
|
24 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
25 |
+
]
|
26 |
+
|
27 |
+
# YOLOv5 v6.0 head
|
28 |
+
head:
|
29 |
+
[[-1, 1, Conv, [512, 1, 1]],
|
30 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
31 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
32 |
+
[-1, 3, C3, [512, False]], # 13
|
33 |
+
|
34 |
+
[-1, 1, Conv, [256, 1, 1]],
|
35 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
36 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
37 |
+
[-1, 3, C3, [256, False]], # 17 (P3/8-small)
|
38 |
+
|
39 |
+
[-1, 1, Conv, [256, 3, 2]],
|
40 |
+
[[-1, 14], 1, Concat, [1]], # cat head P4
|
41 |
+
[-1, 3, C3, [512, False]], # 20 (P4/16-medium)
|
42 |
+
|
43 |
+
[-1, 1, Conv, [512, 3, 2]],
|
44 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
45 |
+
[-1, 3, C3, [1024, False]], # 23 (P5/32-large)
|
46 |
+
|
47 |
+
[[17, 20, 23], 1, Segment, [nc, anchors, 32, 256]], # Detect(P3, P4, P5)
|
48 |
+
]
|
models/segment/yolov5n-seg.yaml
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 π by Ultralytics, AGPL-3.0 license
|
2 |
+
|
3 |
+
# Parameters
|
4 |
+
nc: 80 # number of classes
|
5 |
+
depth_multiple: 0.33 # model depth multiple
|
6 |
+
width_multiple: 0.25 # layer channel multiple
|
7 |
+
anchors:
|
8 |
+
- [10,13, 16,30, 33,23] # P3/8
|
9 |
+
- [30,61, 62,45, 59,119] # P4/16
|
10 |
+
- [116,90, 156,198, 373,326] # P5/32
|
11 |
+
|
12 |
+
# YOLOv5 v6.0 backbone
|
13 |
+
backbone:
|
14 |
+
# [from, number, module, args]
|
15 |
+
[[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2
|
16 |
+
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
|
17 |
+
[-1, 3, C3, [128]],
|
18 |
+
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
|
19 |
+
[-1, 6, C3, [256]],
|
20 |
+
[-1, 1, Conv, [512, 3, 2]], # 5-P4/16
|
21 |
+
[-1, 9, C3, [512]],
|
22 |
+
[-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
|
23 |
+
[-1, 3, C3, [1024]],
|
24 |
+
[-1, 1, SPPF, [1024, 5]], # 9
|
25 |
+
]
|
26 |
+
|
27 |
+
# YOLOv5 v6.0 head
|
28 |
+
head:
|
29 |
+
[[-1, 1, Conv, [512, 1, 1]],
|
30 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
31 |
+
[[-1, 6], 1, Concat, [1]], # cat backbone P4
|
32 |
+
[-1, 3, C3, [512, False]], # 13
|
33 |
+
|
34 |
+
[-1, 1, Conv, [256, 1, 1]],
|
35 |
+
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
|
36 |
+
[[-1, 4], 1, Concat, [1]], # cat backbone P3
|
37 |
+
[-1, 3, C3, [256, False]], # 17 (P3/8-small)
|
38 |
+
|
39 |
+
[-1, 1, Conv, [256, 3, 2]],
|
40 |
+
[[-1, 14], 1, Concat, [1]], # cat head P4
|
41 |
+
[-1, 3, C3, [512, False]], # 20 (P4/16-medium)
|
42 |
+
|
43 |
+
[-1, 1, Conv, [512, 3, 2]],
|
44 |
+
[[-1, 10], 1, Concat, [1]], # cat head P5
|
45 |
+
[-1, 3, C3, [1024, False]], # 23 (P5/32-large)
|
46 |
+
|
47 |
+
[[17, 20, 23], 1, Segment, [nc, anchors, 32, 256]], # Detect(P3, P4, P5)
|
48 |
+
]
|