Spaces:
Runtime error
Runtime error
ossaili
commited on
Commit
•
9b43cf7
1
Parent(s):
080c67f
na
Browse files- app.py +193 -0
- assets/bauhaus.jpg +0 -0
- assets/frank_gehry.jpg +0 -0
- assets/pyramid.jpg +0 -0
- models/model_weights_27_styles.pth +3 -0
- network.txt +1855 -0
- requirements.txt +74 -0
- utils/__init__.py +0 -0
- utils/__pycache__/__init__.cpython-310.pyc +0 -0
- utils/__pycache__/imshow.cpython-310.pyc +0 -0
- utils/__pycache__/save_load.cpython-310.pyc +0 -0
- utils/__pycache__/utils.cpython-310.pyc +0 -0
- utils/imshow.py +18 -0
- utils/save_load.py +10 -0
app.py
CHANGED
@@ -0,0 +1,193 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import sys
|
2 |
+
import PIL
|
3 |
+
import cv2
|
4 |
+
import torch
|
5 |
+
import torchvision
|
6 |
+
import torch.nn as nn
|
7 |
+
from utils.save_load import load_model
|
8 |
+
import gradio as gr
|
9 |
+
from PIL import Image
|
10 |
+
from torchvision import transforms
|
11 |
+
import gradio as gr
|
12 |
+
from pytorch_grad_cam import GradCAM, AblationCAM, FullGrad, EigenGradCAM, LayerCAM
|
13 |
+
from pytorch_grad_cam.utils.image import show_cam_on_image
|
14 |
+
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
|
15 |
+
from pytorch_grad_cam import DeepFeatureFactorization
|
16 |
+
from pytorch_grad_cam.utils.image import show_cam_on_image, preprocess_image, deprocess_image
|
17 |
+
import numpy as np
|
18 |
+
from typing import List
|
19 |
+
from matplotlib import pyplot as plt
|
20 |
+
from matplotlib.lines import Line2D
|
21 |
+
|
22 |
+
labels = [
|
23 |
+
"Achaemenid architecture",
|
24 |
+
"American craftsman style",
|
25 |
+
"American Foursquare architecture",
|
26 |
+
"Ancient Egyptian architecture",
|
27 |
+
"Art Deco architecture",
|
28 |
+
"Art Nouveau architecture",
|
29 |
+
"Baroque architecture",
|
30 |
+
"Bauhaus architecture",
|
31 |
+
"Beaux-Arts architecture",
|
32 |
+
"Brutalism architecture",
|
33 |
+
"Byzantine architecture",
|
34 |
+
"Chicago school architecture",
|
35 |
+
"Colonial architecture",
|
36 |
+
"Deconstructivism",
|
37 |
+
"Edwardian architecture",
|
38 |
+
"Georgian architecture",
|
39 |
+
"Gothic architecture",
|
40 |
+
"Greek Revival architecture",
|
41 |
+
"International style",
|
42 |
+
"Islamic architecture",
|
43 |
+
"Novelty architecture",
|
44 |
+
"Palladian architecture",
|
45 |
+
"Postmodern architecture",
|
46 |
+
"Queen Anne architecture",
|
47 |
+
"Romanesque architecture",
|
48 |
+
"Russian Revival architecture",
|
49 |
+
"Tudor Revival architecture"
|
50 |
+
]
|
51 |
+
|
52 |
+
print(len(labels))
|
53 |
+
model = torchvision.models.efficientnet_v2_l()
|
54 |
+
|
55 |
+
model.classifier = nn.Sequential(
|
56 |
+
nn.Dropout(p=0.4, inplace=True),
|
57 |
+
nn.Linear(1280, len(labels), bias=True)
|
58 |
+
)
|
59 |
+
|
60 |
+
load_model(model)
|
61 |
+
|
62 |
+
|
63 |
+
target_layers = model.features[-1]
|
64 |
+
classifier = model.classifier
|
65 |
+
cam = LayerCAM(model=model, target_layers=target_layers, use_cuda=False)
|
66 |
+
dff = DeepFeatureFactorization(
|
67 |
+
model=model, target_layer=target_layers, computation_on_concepts=classifier)
|
68 |
+
|
69 |
+
|
70 |
+
def show_factorization_on_image(img: np.ndarray,
|
71 |
+
explanations: np.ndarray,
|
72 |
+
colors: List[np.ndarray] = None,
|
73 |
+
image_weight: float = 0.5,
|
74 |
+
concept_labels: List = None) -> np.ndarray:
|
75 |
+
n_components = explanations.shape[0]
|
76 |
+
if colors is None:
|
77 |
+
# taken from https://github.com/edocollins/DFF/blob/master/utils.py
|
78 |
+
_cmap = plt.cm.get_cmap('gist_rainbow')
|
79 |
+
colors = [
|
80 |
+
np.array(
|
81 |
+
_cmap(i)) for i in np.arange(
|
82 |
+
0,
|
83 |
+
1,
|
84 |
+
1.0 /
|
85 |
+
n_components)]
|
86 |
+
concept_per_pixel = explanations.argmax(axis=0)
|
87 |
+
masks = []
|
88 |
+
for i in range(n_components):
|
89 |
+
mask = np.zeros(shape=(img.shape[0], img.shape[1], 3))
|
90 |
+
mask[:, :, :] = colors[i][:3]
|
91 |
+
explanation = explanations[i]
|
92 |
+
explanation[concept_per_pixel != i] = 0
|
93 |
+
mask = np.uint8(mask * 255)
|
94 |
+
mask = cv2.cvtColor(mask, cv2.COLOR_RGB2HSV)
|
95 |
+
mask[:, :, 2] = np.uint8(255 * explanation)
|
96 |
+
mask = cv2.cvtColor(mask, cv2.COLOR_HSV2RGB)
|
97 |
+
mask = np.float32(mask) / 255
|
98 |
+
masks.append(mask)
|
99 |
+
|
100 |
+
mask = np.sum(np.float32(masks), axis=0)
|
101 |
+
result = img * image_weight + mask * (1 - image_weight)
|
102 |
+
result = np.uint8(result * 255)
|
103 |
+
|
104 |
+
if concept_labels is not None:
|
105 |
+
px = 1 / plt.rcParams['figure.dpi'] # pixel in inches
|
106 |
+
fig = plt.figure(figsize=(result.shape[1] * px, result.shape[0] * px))
|
107 |
+
plt.rcParams['legend.fontsize'] = 6 * result.shape[0] / 256
|
108 |
+
lw = 5 * result.shape[0] / 256
|
109 |
+
lines = [Line2D([0], [0], color=colors[i], lw=lw)
|
110 |
+
for i in range(n_components)]
|
111 |
+
plt.legend(lines,
|
112 |
+
concept_labels,
|
113 |
+
|
114 |
+
fancybox=False,
|
115 |
+
shadow=False,
|
116 |
+
frameon=False,
|
117 |
+
loc="center")
|
118 |
+
|
119 |
+
plt.tight_layout(pad=0, w_pad=0, h_pad=0)
|
120 |
+
plt.axis('off')
|
121 |
+
fig.canvas.draw()
|
122 |
+
data = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)
|
123 |
+
plt.close(fig=fig)
|
124 |
+
data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
|
125 |
+
data = cv2.resize(data, (result.shape[1], result.shape[0]))
|
126 |
+
result = np.vstack((result, data))
|
127 |
+
return result
|
128 |
+
|
129 |
+
|
130 |
+
def create_labels(concept_scores, top_k=2):
|
131 |
+
""" Create a list with the image-net category names of the top scoring categories"""
|
132 |
+
concept_categories = np.argsort(concept_scores, axis=1)[:, ::-1][:, :top_k]
|
133 |
+
concept_labels_topk = []
|
134 |
+
for concept_index in range(concept_categories.shape[0]):
|
135 |
+
categories = concept_categories[concept_index, :]
|
136 |
+
concept_labels = []
|
137 |
+
for category in categories:
|
138 |
+
score = concept_scores[concept_index, category]
|
139 |
+
label = f"{labels[category].split(',')[0]}:{score*100:.2f}%"
|
140 |
+
concept_labels.append(label)
|
141 |
+
concept_labels_topk.append("\n".join(concept_labels))
|
142 |
+
return concept_labels_topk
|
143 |
+
|
144 |
+
|
145 |
+
def predict(rgb_img, top_k):
|
146 |
+
print(top_k)
|
147 |
+
inp_01 = transforms.Compose(
|
148 |
+
[
|
149 |
+
transforms.ToTensor(),
|
150 |
+
transforms.Normalize([0.4937, 0.5060, 0.5030], [
|
151 |
+
0.2705, 0.2653, 0.2998]),
|
152 |
+
transforms.Resize((224, 224)),
|
153 |
+
])(rgb_img)
|
154 |
+
|
155 |
+
model.eval()
|
156 |
+
with torch.no_grad():
|
157 |
+
prediction = torch.nn.functional.softmax(
|
158 |
+
model(inp_01.unsqueeze(0))[0], dim=0)
|
159 |
+
confidences = {labels[i]: float(prediction[i])
|
160 |
+
for i in range(len(labels))}
|
161 |
+
|
162 |
+
concepts, batch_explanations, concept_outputs = dff(
|
163 |
+
inp_01.unsqueeze(0), 5)
|
164 |
+
|
165 |
+
concept_outputs = torch.softmax(
|
166 |
+
torch.from_numpy(concept_outputs), axis=-1).numpy()
|
167 |
+
concept_label_strings = create_labels(concept_outputs, top_k=top_k)
|
168 |
+
|
169 |
+
print(inp_01.shape)
|
170 |
+
print(batch_explanations[0].shape)
|
171 |
+
res = cv2.resize(np.transpose(
|
172 |
+
batch_explanations[0], (1, 2, 0)), (rgb_img.size[0], rgb_img.size[1]))
|
173 |
+
res = np.transpose(res, (2, 0, 1))
|
174 |
+
print(res.shape)
|
175 |
+
|
176 |
+
visualization_01 = show_factorization_on_image(np.float32(rgb_img)/255.0,
|
177 |
+
res,
|
178 |
+
image_weight=0.3,
|
179 |
+
concept_labels=concept_label_strings)
|
180 |
+
|
181 |
+
return confidences, visualization_01,
|
182 |
+
|
183 |
+
|
184 |
+
gr.Interface(fn=predict,
|
185 |
+
inputs=[gr.Image(type="pil"), gr.Slider(
|
186 |
+
minimum=1, maximum=4, label="Number of top results", step=1)],
|
187 |
+
outputs=[gr.Label(num_top_classes=5), "image"],
|
188 |
+
examples=[["./assets/bauhaus.jpg", 1],
|
189 |
+
["./assets/frank_gehry.jpg", 2], ["./assets/pyramid.jpg", 3]]
|
190 |
+
).launch()
|
191 |
+
|
192 |
+
|
193 |
+
# examples=["./assets/bauhaus.jpg", "./assets/frank_gehry.jpg", "./assets/pyramid.jpg"]
|
assets/bauhaus.jpg
ADDED
assets/frank_gehry.jpg
ADDED
assets/pyramid.jpg
ADDED
models/model_weights_27_styles.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:58ca956f118139d5e28e3181e80cd5d408f1a090656c9dba0c58dc4e260619c7
|
3 |
+
size 471688845
|
network.txt
ADDED
@@ -0,0 +1,1855 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
EfficientNet(
|
2 |
+
(features): Sequential(
|
3 |
+
(0): Conv2dNormActivation(
|
4 |
+
(0): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
|
5 |
+
(1): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
6 |
+
(2): SiLU(inplace=True)
|
7 |
+
)
|
8 |
+
(1): Sequential(
|
9 |
+
(0): FusedMBConv(
|
10 |
+
(block): Sequential(
|
11 |
+
(0): Conv2dNormActivation(
|
12 |
+
(0): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
13 |
+
(1): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
14 |
+
(2): SiLU(inplace=True)
|
15 |
+
)
|
16 |
+
)
|
17 |
+
(stochastic_depth): StochasticDepth(p=0.0, mode=row)
|
18 |
+
)
|
19 |
+
(1): FusedMBConv(
|
20 |
+
(block): Sequential(
|
21 |
+
(0): Conv2dNormActivation(
|
22 |
+
(0): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
23 |
+
(1): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
24 |
+
(2): SiLU(inplace=True)
|
25 |
+
)
|
26 |
+
)
|
27 |
+
(stochastic_depth): StochasticDepth(p=0.002531645569620253, mode=row)
|
28 |
+
)
|
29 |
+
(2): FusedMBConv(
|
30 |
+
(block): Sequential(
|
31 |
+
(0): Conv2dNormActivation(
|
32 |
+
(0): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
33 |
+
(1): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
34 |
+
(2): SiLU(inplace=True)
|
35 |
+
)
|
36 |
+
)
|
37 |
+
(stochastic_depth): StochasticDepth(p=0.005063291139240506, mode=row)
|
38 |
+
)
|
39 |
+
(3): FusedMBConv(
|
40 |
+
(block): Sequential(
|
41 |
+
(0): Conv2dNormActivation(
|
42 |
+
(0): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
43 |
+
(1): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
44 |
+
(2): SiLU(inplace=True)
|
45 |
+
)
|
46 |
+
)
|
47 |
+
(stochastic_depth): StochasticDepth(p=0.007594936708860761, mode=row)
|
48 |
+
)
|
49 |
+
)
|
50 |
+
(2): Sequential(
|
51 |
+
(0): FusedMBConv(
|
52 |
+
(block): Sequential(
|
53 |
+
(0): Conv2dNormActivation(
|
54 |
+
(0): Conv2d(32, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
|
55 |
+
(1): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
56 |
+
(2): SiLU(inplace=True)
|
57 |
+
)
|
58 |
+
(1): Conv2dNormActivation(
|
59 |
+
(0): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
60 |
+
(1): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
61 |
+
)
|
62 |
+
)
|
63 |
+
(stochastic_depth): StochasticDepth(p=0.010126582278481013, mode=row)
|
64 |
+
)
|
65 |
+
(1): FusedMBConv(
|
66 |
+
(block): Sequential(
|
67 |
+
(0): Conv2dNormActivation(
|
68 |
+
(0): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
69 |
+
(1): BatchNorm2d(256, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
70 |
+
(2): SiLU(inplace=True)
|
71 |
+
)
|
72 |
+
(1): Conv2dNormActivation(
|
73 |
+
(0): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
74 |
+
(1): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
75 |
+
)
|
76 |
+
)
|
77 |
+
(stochastic_depth): StochasticDepth(p=0.012658227848101266, mode=row)
|
78 |
+
)
|
79 |
+
(2): FusedMBConv(
|
80 |
+
(block): Sequential(
|
81 |
+
(0): Conv2dNormActivation(
|
82 |
+
(0): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
83 |
+
(1): BatchNorm2d(256, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
84 |
+
(2): SiLU(inplace=True)
|
85 |
+
)
|
86 |
+
(1): Conv2dNormActivation(
|
87 |
+
(0): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
88 |
+
(1): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
89 |
+
)
|
90 |
+
)
|
91 |
+
(stochastic_depth): StochasticDepth(p=0.015189873417721522, mode=row)
|
92 |
+
)
|
93 |
+
(3): FusedMBConv(
|
94 |
+
(block): Sequential(
|
95 |
+
(0): Conv2dNormActivation(
|
96 |
+
(0): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
97 |
+
(1): BatchNorm2d(256, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
98 |
+
(2): SiLU(inplace=True)
|
99 |
+
)
|
100 |
+
(1): Conv2dNormActivation(
|
101 |
+
(0): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
102 |
+
(1): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
103 |
+
)
|
104 |
+
)
|
105 |
+
(stochastic_depth): StochasticDepth(p=0.017721518987341773, mode=row)
|
106 |
+
)
|
107 |
+
(4): FusedMBConv(
|
108 |
+
(block): Sequential(
|
109 |
+
(0): Conv2dNormActivation(
|
110 |
+
(0): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
111 |
+
(1): BatchNorm2d(256, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
112 |
+
(2): SiLU(inplace=True)
|
113 |
+
)
|
114 |
+
(1): Conv2dNormActivation(
|
115 |
+
(0): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
116 |
+
(1): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
117 |
+
)
|
118 |
+
)
|
119 |
+
(stochastic_depth): StochasticDepth(p=0.020253164556962026, mode=row)
|
120 |
+
)
|
121 |
+
(5): FusedMBConv(
|
122 |
+
(block): Sequential(
|
123 |
+
(0): Conv2dNormActivation(
|
124 |
+
(0): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
125 |
+
(1): BatchNorm2d(256, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
126 |
+
(2): SiLU(inplace=True)
|
127 |
+
)
|
128 |
+
(1): Conv2dNormActivation(
|
129 |
+
(0): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
130 |
+
(1): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
131 |
+
)
|
132 |
+
)
|
133 |
+
(stochastic_depth): StochasticDepth(p=0.02278481012658228, mode=row)
|
134 |
+
)
|
135 |
+
(6): FusedMBConv(
|
136 |
+
(block): Sequential(
|
137 |
+
(0): Conv2dNormActivation(
|
138 |
+
(0): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
139 |
+
(1): BatchNorm2d(256, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
140 |
+
(2): SiLU(inplace=True)
|
141 |
+
)
|
142 |
+
(1): Conv2dNormActivation(
|
143 |
+
(0): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
144 |
+
(1): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
145 |
+
)
|
146 |
+
)
|
147 |
+
(stochastic_depth): StochasticDepth(p=0.02531645569620253, mode=row)
|
148 |
+
)
|
149 |
+
)
|
150 |
+
(3): Sequential(
|
151 |
+
(0): FusedMBConv(
|
152 |
+
(block): Sequential(
|
153 |
+
(0): Conv2dNormActivation(
|
154 |
+
(0): Conv2d(64, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
|
155 |
+
(1): BatchNorm2d(256, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
156 |
+
(2): SiLU(inplace=True)
|
157 |
+
)
|
158 |
+
(1): Conv2dNormActivation(
|
159 |
+
(0): Conv2d(256, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
160 |
+
(1): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
161 |
+
)
|
162 |
+
)
|
163 |
+
(stochastic_depth): StochasticDepth(p=0.027848101265822787, mode=row)
|
164 |
+
)
|
165 |
+
(1): FusedMBConv(
|
166 |
+
(block): Sequential(
|
167 |
+
(0): Conv2dNormActivation(
|
168 |
+
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
169 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
170 |
+
(2): SiLU(inplace=True)
|
171 |
+
)
|
172 |
+
(1): Conv2dNormActivation(
|
173 |
+
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
174 |
+
(1): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
175 |
+
)
|
176 |
+
)
|
177 |
+
(stochastic_depth): StochasticDepth(p=0.030379746835443044, mode=row)
|
178 |
+
)
|
179 |
+
(2): FusedMBConv(
|
180 |
+
(block): Sequential(
|
181 |
+
(0): Conv2dNormActivation(
|
182 |
+
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
183 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
184 |
+
(2): SiLU(inplace=True)
|
185 |
+
)
|
186 |
+
(1): Conv2dNormActivation(
|
187 |
+
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
188 |
+
(1): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
189 |
+
)
|
190 |
+
)
|
191 |
+
(stochastic_depth): StochasticDepth(p=0.03291139240506329, mode=row)
|
192 |
+
)
|
193 |
+
(3): FusedMBConv(
|
194 |
+
(block): Sequential(
|
195 |
+
(0): Conv2dNormActivation(
|
196 |
+
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
197 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
198 |
+
(2): SiLU(inplace=True)
|
199 |
+
)
|
200 |
+
(1): Conv2dNormActivation(
|
201 |
+
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
202 |
+
(1): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
203 |
+
)
|
204 |
+
)
|
205 |
+
(stochastic_depth): StochasticDepth(p=0.035443037974683546, mode=row)
|
206 |
+
)
|
207 |
+
(4): FusedMBConv(
|
208 |
+
(block): Sequential(
|
209 |
+
(0): Conv2dNormActivation(
|
210 |
+
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
211 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
212 |
+
(2): SiLU(inplace=True)
|
213 |
+
)
|
214 |
+
(1): Conv2dNormActivation(
|
215 |
+
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
216 |
+
(1): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
217 |
+
)
|
218 |
+
)
|
219 |
+
(stochastic_depth): StochasticDepth(p=0.0379746835443038, mode=row)
|
220 |
+
)
|
221 |
+
(5): FusedMBConv(
|
222 |
+
(block): Sequential(
|
223 |
+
(0): Conv2dNormActivation(
|
224 |
+
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
225 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
226 |
+
(2): SiLU(inplace=True)
|
227 |
+
)
|
228 |
+
(1): Conv2dNormActivation(
|
229 |
+
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
230 |
+
(1): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
231 |
+
)
|
232 |
+
)
|
233 |
+
(stochastic_depth): StochasticDepth(p=0.04050632911392405, mode=row)
|
234 |
+
)
|
235 |
+
(6): FusedMBConv(
|
236 |
+
(block): Sequential(
|
237 |
+
(0): Conv2dNormActivation(
|
238 |
+
(0): Conv2d(96, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
|
239 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
240 |
+
(2): SiLU(inplace=True)
|
241 |
+
)
|
242 |
+
(1): Conv2dNormActivation(
|
243 |
+
(0): Conv2d(384, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
244 |
+
(1): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
245 |
+
)
|
246 |
+
)
|
247 |
+
(stochastic_depth): StochasticDepth(p=0.04303797468354431, mode=row)
|
248 |
+
)
|
249 |
+
)
|
250 |
+
(4): Sequential(
|
251 |
+
(0): MBConv(
|
252 |
+
(block): Sequential(
|
253 |
+
(0): Conv2dNormActivation(
|
254 |
+
(0): Conv2d(96, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
255 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
256 |
+
(2): SiLU(inplace=True)
|
257 |
+
)
|
258 |
+
(1): Conv2dNormActivation(
|
259 |
+
(0): Conv2d(384, 384, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=384, bias=False)
|
260 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
261 |
+
(2): SiLU(inplace=True)
|
262 |
+
)
|
263 |
+
(2): SqueezeExcitation(
|
264 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
265 |
+
(fc1): Conv2d(384, 24, kernel_size=(1, 1), stride=(1, 1))
|
266 |
+
(fc2): Conv2d(24, 384, kernel_size=(1, 1), stride=(1, 1))
|
267 |
+
(activation): SiLU(inplace=True)
|
268 |
+
(scale_activation): Sigmoid()
|
269 |
+
)
|
270 |
+
(3): Conv2dNormActivation(
|
271 |
+
(0): Conv2d(384, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
272 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
273 |
+
)
|
274 |
+
)
|
275 |
+
(stochastic_depth): StochasticDepth(p=0.04556962025316456, mode=row)
|
276 |
+
)
|
277 |
+
(1): MBConv(
|
278 |
+
(block): Sequential(
|
279 |
+
(0): Conv2dNormActivation(
|
280 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
281 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
282 |
+
(2): SiLU(inplace=True)
|
283 |
+
)
|
284 |
+
(1): Conv2dNormActivation(
|
285 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
286 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
287 |
+
(2): SiLU(inplace=True)
|
288 |
+
)
|
289 |
+
(2): SqueezeExcitation(
|
290 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
291 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
292 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
293 |
+
(activation): SiLU(inplace=True)
|
294 |
+
(scale_activation): Sigmoid()
|
295 |
+
)
|
296 |
+
(3): Conv2dNormActivation(
|
297 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
298 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
299 |
+
)
|
300 |
+
)
|
301 |
+
(stochastic_depth): StochasticDepth(p=0.04810126582278482, mode=row)
|
302 |
+
)
|
303 |
+
(2): MBConv(
|
304 |
+
(block): Sequential(
|
305 |
+
(0): Conv2dNormActivation(
|
306 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
307 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
308 |
+
(2): SiLU(inplace=True)
|
309 |
+
)
|
310 |
+
(1): Conv2dNormActivation(
|
311 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
312 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
313 |
+
(2): SiLU(inplace=True)
|
314 |
+
)
|
315 |
+
(2): SqueezeExcitation(
|
316 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
317 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
318 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
319 |
+
(activation): SiLU(inplace=True)
|
320 |
+
(scale_activation): Sigmoid()
|
321 |
+
)
|
322 |
+
(3): Conv2dNormActivation(
|
323 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
324 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
325 |
+
)
|
326 |
+
)
|
327 |
+
(stochastic_depth): StochasticDepth(p=0.05063291139240506, mode=row)
|
328 |
+
)
|
329 |
+
(3): MBConv(
|
330 |
+
(block): Sequential(
|
331 |
+
(0): Conv2dNormActivation(
|
332 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
333 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
334 |
+
(2): SiLU(inplace=True)
|
335 |
+
)
|
336 |
+
(1): Conv2dNormActivation(
|
337 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
338 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
339 |
+
(2): SiLU(inplace=True)
|
340 |
+
)
|
341 |
+
(2): SqueezeExcitation(
|
342 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
343 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
344 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
345 |
+
(activation): SiLU(inplace=True)
|
346 |
+
(scale_activation): Sigmoid()
|
347 |
+
)
|
348 |
+
(3): Conv2dNormActivation(
|
349 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
350 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
351 |
+
)
|
352 |
+
)
|
353 |
+
(stochastic_depth): StochasticDepth(p=0.053164556962025315, mode=row)
|
354 |
+
)
|
355 |
+
(4): MBConv(
|
356 |
+
(block): Sequential(
|
357 |
+
(0): Conv2dNormActivation(
|
358 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
359 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
360 |
+
(2): SiLU(inplace=True)
|
361 |
+
)
|
362 |
+
(1): Conv2dNormActivation(
|
363 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
364 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
365 |
+
(2): SiLU(inplace=True)
|
366 |
+
)
|
367 |
+
(2): SqueezeExcitation(
|
368 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
369 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
370 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
371 |
+
(activation): SiLU(inplace=True)
|
372 |
+
(scale_activation): Sigmoid()
|
373 |
+
)
|
374 |
+
(3): Conv2dNormActivation(
|
375 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
376 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
377 |
+
)
|
378 |
+
)
|
379 |
+
(stochastic_depth): StochasticDepth(p=0.055696202531645575, mode=row)
|
380 |
+
)
|
381 |
+
(5): MBConv(
|
382 |
+
(block): Sequential(
|
383 |
+
(0): Conv2dNormActivation(
|
384 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
385 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
386 |
+
(2): SiLU(inplace=True)
|
387 |
+
)
|
388 |
+
(1): Conv2dNormActivation(
|
389 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
390 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
391 |
+
(2): SiLU(inplace=True)
|
392 |
+
)
|
393 |
+
(2): SqueezeExcitation(
|
394 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
395 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
396 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
397 |
+
(activation): SiLU(inplace=True)
|
398 |
+
(scale_activation): Sigmoid()
|
399 |
+
)
|
400 |
+
(3): Conv2dNormActivation(
|
401 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
402 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
403 |
+
)
|
404 |
+
)
|
405 |
+
(stochastic_depth): StochasticDepth(p=0.05822784810126583, mode=row)
|
406 |
+
)
|
407 |
+
(6): MBConv(
|
408 |
+
(block): Sequential(
|
409 |
+
(0): Conv2dNormActivation(
|
410 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
411 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
412 |
+
(2): SiLU(inplace=True)
|
413 |
+
)
|
414 |
+
(1): Conv2dNormActivation(
|
415 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
416 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
417 |
+
(2): SiLU(inplace=True)
|
418 |
+
)
|
419 |
+
(2): SqueezeExcitation(
|
420 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
421 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
422 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
423 |
+
(activation): SiLU(inplace=True)
|
424 |
+
(scale_activation): Sigmoid()
|
425 |
+
)
|
426 |
+
(3): Conv2dNormActivation(
|
427 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
428 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
429 |
+
)
|
430 |
+
)
|
431 |
+
(stochastic_depth): StochasticDepth(p=0.06075949367088609, mode=row)
|
432 |
+
)
|
433 |
+
(7): MBConv(
|
434 |
+
(block): Sequential(
|
435 |
+
(0): Conv2dNormActivation(
|
436 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
437 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
438 |
+
(2): SiLU(inplace=True)
|
439 |
+
)
|
440 |
+
(1): Conv2dNormActivation(
|
441 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
442 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
443 |
+
(2): SiLU(inplace=True)
|
444 |
+
)
|
445 |
+
(2): SqueezeExcitation(
|
446 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
447 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
448 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
449 |
+
(activation): SiLU(inplace=True)
|
450 |
+
(scale_activation): Sigmoid()
|
451 |
+
)
|
452 |
+
(3): Conv2dNormActivation(
|
453 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
454 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
455 |
+
)
|
456 |
+
)
|
457 |
+
(stochastic_depth): StochasticDepth(p=0.06329113924050633, mode=row)
|
458 |
+
)
|
459 |
+
(8): MBConv(
|
460 |
+
(block): Sequential(
|
461 |
+
(0): Conv2dNormActivation(
|
462 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
463 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
464 |
+
(2): SiLU(inplace=True)
|
465 |
+
)
|
466 |
+
(1): Conv2dNormActivation(
|
467 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
468 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
469 |
+
(2): SiLU(inplace=True)
|
470 |
+
)
|
471 |
+
(2): SqueezeExcitation(
|
472 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
473 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
474 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
475 |
+
(activation): SiLU(inplace=True)
|
476 |
+
(scale_activation): Sigmoid()
|
477 |
+
)
|
478 |
+
(3): Conv2dNormActivation(
|
479 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
480 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
481 |
+
)
|
482 |
+
)
|
483 |
+
(stochastic_depth): StochasticDepth(p=0.06582278481012659, mode=row)
|
484 |
+
)
|
485 |
+
(9): MBConv(
|
486 |
+
(block): Sequential(
|
487 |
+
(0): Conv2dNormActivation(
|
488 |
+
(0): Conv2d(192, 768, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
489 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
490 |
+
(2): SiLU(inplace=True)
|
491 |
+
)
|
492 |
+
(1): Conv2dNormActivation(
|
493 |
+
(0): Conv2d(768, 768, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=768, bias=False)
|
494 |
+
(1): BatchNorm2d(768, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
495 |
+
(2): SiLU(inplace=True)
|
496 |
+
)
|
497 |
+
(2): SqueezeExcitation(
|
498 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
499 |
+
(fc1): Conv2d(768, 48, kernel_size=(1, 1), stride=(1, 1))
|
500 |
+
(fc2): Conv2d(48, 768, kernel_size=(1, 1), stride=(1, 1))
|
501 |
+
(activation): SiLU(inplace=True)
|
502 |
+
(scale_activation): Sigmoid()
|
503 |
+
)
|
504 |
+
(3): Conv2dNormActivation(
|
505 |
+
(0): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
506 |
+
(1): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
507 |
+
)
|
508 |
+
)
|
509 |
+
(stochastic_depth): StochasticDepth(p=0.06835443037974684, mode=row)
|
510 |
+
)
|
511 |
+
)
|
512 |
+
(5): Sequential(
|
513 |
+
(0): MBConv(
|
514 |
+
(block): Sequential(
|
515 |
+
(0): Conv2dNormActivation(
|
516 |
+
(0): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
517 |
+
(1): BatchNorm2d(1152, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
518 |
+
(2): SiLU(inplace=True)
|
519 |
+
)
|
520 |
+
(1): Conv2dNormActivation(
|
521 |
+
(0): Conv2d(1152, 1152, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1152, bias=False)
|
522 |
+
(1): BatchNorm2d(1152, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
523 |
+
(2): SiLU(inplace=True)
|
524 |
+
)
|
525 |
+
(2): SqueezeExcitation(
|
526 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
527 |
+
(fc1): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))
|
528 |
+
(fc2): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))
|
529 |
+
(activation): SiLU(inplace=True)
|
530 |
+
(scale_activation): Sigmoid()
|
531 |
+
)
|
532 |
+
(3): Conv2dNormActivation(
|
533 |
+
(0): Conv2d(1152, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
534 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
535 |
+
)
|
536 |
+
)
|
537 |
+
(stochastic_depth): StochasticDepth(p=0.07088607594936709, mode=row)
|
538 |
+
)
|
539 |
+
(1): MBConv(
|
540 |
+
(block): Sequential(
|
541 |
+
(0): Conv2dNormActivation(
|
542 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
543 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
544 |
+
(2): SiLU(inplace=True)
|
545 |
+
)
|
546 |
+
(1): Conv2dNormActivation(
|
547 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
548 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
549 |
+
(2): SiLU(inplace=True)
|
550 |
+
)
|
551 |
+
(2): SqueezeExcitation(
|
552 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
553 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
554 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
555 |
+
(activation): SiLU(inplace=True)
|
556 |
+
(scale_activation): Sigmoid()
|
557 |
+
)
|
558 |
+
(3): Conv2dNormActivation(
|
559 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
560 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
561 |
+
)
|
562 |
+
)
|
563 |
+
(stochastic_depth): StochasticDepth(p=0.07341772151898734, mode=row)
|
564 |
+
)
|
565 |
+
(2): MBConv(
|
566 |
+
(block): Sequential(
|
567 |
+
(0): Conv2dNormActivation(
|
568 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
569 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
570 |
+
(2): SiLU(inplace=True)
|
571 |
+
)
|
572 |
+
(1): Conv2dNormActivation(
|
573 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
574 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
575 |
+
(2): SiLU(inplace=True)
|
576 |
+
)
|
577 |
+
(2): SqueezeExcitation(
|
578 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
579 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
580 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
581 |
+
(activation): SiLU(inplace=True)
|
582 |
+
(scale_activation): Sigmoid()
|
583 |
+
)
|
584 |
+
(3): Conv2dNormActivation(
|
585 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
586 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
587 |
+
)
|
588 |
+
)
|
589 |
+
(stochastic_depth): StochasticDepth(p=0.0759493670886076, mode=row)
|
590 |
+
)
|
591 |
+
(3): MBConv(
|
592 |
+
(block): Sequential(
|
593 |
+
(0): Conv2dNormActivation(
|
594 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
595 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
596 |
+
(2): SiLU(inplace=True)
|
597 |
+
)
|
598 |
+
(1): Conv2dNormActivation(
|
599 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
600 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
601 |
+
(2): SiLU(inplace=True)
|
602 |
+
)
|
603 |
+
(2): SqueezeExcitation(
|
604 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
605 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
606 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
607 |
+
(activation): SiLU(inplace=True)
|
608 |
+
(scale_activation): Sigmoid()
|
609 |
+
)
|
610 |
+
(3): Conv2dNormActivation(
|
611 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
612 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
613 |
+
)
|
614 |
+
)
|
615 |
+
(stochastic_depth): StochasticDepth(p=0.07848101265822785, mode=row)
|
616 |
+
)
|
617 |
+
(4): MBConv(
|
618 |
+
(block): Sequential(
|
619 |
+
(0): Conv2dNormActivation(
|
620 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
621 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
622 |
+
(2): SiLU(inplace=True)
|
623 |
+
)
|
624 |
+
(1): Conv2dNormActivation(
|
625 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
626 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
627 |
+
(2): SiLU(inplace=True)
|
628 |
+
)
|
629 |
+
(2): SqueezeExcitation(
|
630 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
631 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
632 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
633 |
+
(activation): SiLU(inplace=True)
|
634 |
+
(scale_activation): Sigmoid()
|
635 |
+
)
|
636 |
+
(3): Conv2dNormActivation(
|
637 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
638 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
639 |
+
)
|
640 |
+
)
|
641 |
+
(stochastic_depth): StochasticDepth(p=0.0810126582278481, mode=row)
|
642 |
+
)
|
643 |
+
(5): MBConv(
|
644 |
+
(block): Sequential(
|
645 |
+
(0): Conv2dNormActivation(
|
646 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
647 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
648 |
+
(2): SiLU(inplace=True)
|
649 |
+
)
|
650 |
+
(1): Conv2dNormActivation(
|
651 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
652 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
653 |
+
(2): SiLU(inplace=True)
|
654 |
+
)
|
655 |
+
(2): SqueezeExcitation(
|
656 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
657 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
658 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
659 |
+
(activation): SiLU(inplace=True)
|
660 |
+
(scale_activation): Sigmoid()
|
661 |
+
)
|
662 |
+
(3): Conv2dNormActivation(
|
663 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
664 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
665 |
+
)
|
666 |
+
)
|
667 |
+
(stochastic_depth): StochasticDepth(p=0.08354430379746836, mode=row)
|
668 |
+
)
|
669 |
+
(6): MBConv(
|
670 |
+
(block): Sequential(
|
671 |
+
(0): Conv2dNormActivation(
|
672 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
673 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
674 |
+
(2): SiLU(inplace=True)
|
675 |
+
)
|
676 |
+
(1): Conv2dNormActivation(
|
677 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
678 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
679 |
+
(2): SiLU(inplace=True)
|
680 |
+
)
|
681 |
+
(2): SqueezeExcitation(
|
682 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
683 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
684 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
685 |
+
(activation): SiLU(inplace=True)
|
686 |
+
(scale_activation): Sigmoid()
|
687 |
+
)
|
688 |
+
(3): Conv2dNormActivation(
|
689 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
690 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
691 |
+
)
|
692 |
+
)
|
693 |
+
(stochastic_depth): StochasticDepth(p=0.08607594936708862, mode=row)
|
694 |
+
)
|
695 |
+
(7): MBConv(
|
696 |
+
(block): Sequential(
|
697 |
+
(0): Conv2dNormActivation(
|
698 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
699 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
700 |
+
(2): SiLU(inplace=True)
|
701 |
+
)
|
702 |
+
(1): Conv2dNormActivation(
|
703 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
704 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
705 |
+
(2): SiLU(inplace=True)
|
706 |
+
)
|
707 |
+
(2): SqueezeExcitation(
|
708 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
709 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
710 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
711 |
+
(activation): SiLU(inplace=True)
|
712 |
+
(scale_activation): Sigmoid()
|
713 |
+
)
|
714 |
+
(3): Conv2dNormActivation(
|
715 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
716 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
717 |
+
)
|
718 |
+
)
|
719 |
+
(stochastic_depth): StochasticDepth(p=0.08860759493670886, mode=row)
|
720 |
+
)
|
721 |
+
(8): MBConv(
|
722 |
+
(block): Sequential(
|
723 |
+
(0): Conv2dNormActivation(
|
724 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
725 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
726 |
+
(2): SiLU(inplace=True)
|
727 |
+
)
|
728 |
+
(1): Conv2dNormActivation(
|
729 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
730 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
731 |
+
(2): SiLU(inplace=True)
|
732 |
+
)
|
733 |
+
(2): SqueezeExcitation(
|
734 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
735 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
736 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
737 |
+
(activation): SiLU(inplace=True)
|
738 |
+
(scale_activation): Sigmoid()
|
739 |
+
)
|
740 |
+
(3): Conv2dNormActivation(
|
741 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
742 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
743 |
+
)
|
744 |
+
)
|
745 |
+
(stochastic_depth): StochasticDepth(p=0.09113924050632911, mode=row)
|
746 |
+
)
|
747 |
+
(9): MBConv(
|
748 |
+
(block): Sequential(
|
749 |
+
(0): Conv2dNormActivation(
|
750 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
751 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
752 |
+
(2): SiLU(inplace=True)
|
753 |
+
)
|
754 |
+
(1): Conv2dNormActivation(
|
755 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
756 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
757 |
+
(2): SiLU(inplace=True)
|
758 |
+
)
|
759 |
+
(2): SqueezeExcitation(
|
760 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
761 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
762 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
763 |
+
(activation): SiLU(inplace=True)
|
764 |
+
(scale_activation): Sigmoid()
|
765 |
+
)
|
766 |
+
(3): Conv2dNormActivation(
|
767 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
768 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
769 |
+
)
|
770 |
+
)
|
771 |
+
(stochastic_depth): StochasticDepth(p=0.09367088607594937, mode=row)
|
772 |
+
)
|
773 |
+
(10): MBConv(
|
774 |
+
(block): Sequential(
|
775 |
+
(0): Conv2dNormActivation(
|
776 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
777 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
778 |
+
(2): SiLU(inplace=True)
|
779 |
+
)
|
780 |
+
(1): Conv2dNormActivation(
|
781 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
782 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
783 |
+
(2): SiLU(inplace=True)
|
784 |
+
)
|
785 |
+
(2): SqueezeExcitation(
|
786 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
787 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
788 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
789 |
+
(activation): SiLU(inplace=True)
|
790 |
+
(scale_activation): Sigmoid()
|
791 |
+
)
|
792 |
+
(3): Conv2dNormActivation(
|
793 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
794 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
795 |
+
)
|
796 |
+
)
|
797 |
+
(stochastic_depth): StochasticDepth(p=0.09620253164556963, mode=row)
|
798 |
+
)
|
799 |
+
(11): MBConv(
|
800 |
+
(block): Sequential(
|
801 |
+
(0): Conv2dNormActivation(
|
802 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
803 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
804 |
+
(2): SiLU(inplace=True)
|
805 |
+
)
|
806 |
+
(1): Conv2dNormActivation(
|
807 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
808 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
809 |
+
(2): SiLU(inplace=True)
|
810 |
+
)
|
811 |
+
(2): SqueezeExcitation(
|
812 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
813 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
814 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
815 |
+
(activation): SiLU(inplace=True)
|
816 |
+
(scale_activation): Sigmoid()
|
817 |
+
)
|
818 |
+
(3): Conv2dNormActivation(
|
819 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
820 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
821 |
+
)
|
822 |
+
)
|
823 |
+
(stochastic_depth): StochasticDepth(p=0.09873417721518989, mode=row)
|
824 |
+
)
|
825 |
+
(12): MBConv(
|
826 |
+
(block): Sequential(
|
827 |
+
(0): Conv2dNormActivation(
|
828 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
829 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
830 |
+
(2): SiLU(inplace=True)
|
831 |
+
)
|
832 |
+
(1): Conv2dNormActivation(
|
833 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
834 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
835 |
+
(2): SiLU(inplace=True)
|
836 |
+
)
|
837 |
+
(2): SqueezeExcitation(
|
838 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
839 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
840 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
841 |
+
(activation): SiLU(inplace=True)
|
842 |
+
(scale_activation): Sigmoid()
|
843 |
+
)
|
844 |
+
(3): Conv2dNormActivation(
|
845 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
846 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
847 |
+
)
|
848 |
+
)
|
849 |
+
(stochastic_depth): StochasticDepth(p=0.10126582278481013, mode=row)
|
850 |
+
)
|
851 |
+
(13): MBConv(
|
852 |
+
(block): Sequential(
|
853 |
+
(0): Conv2dNormActivation(
|
854 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
855 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
856 |
+
(2): SiLU(inplace=True)
|
857 |
+
)
|
858 |
+
(1): Conv2dNormActivation(
|
859 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
860 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
861 |
+
(2): SiLU(inplace=True)
|
862 |
+
)
|
863 |
+
(2): SqueezeExcitation(
|
864 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
865 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
866 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
867 |
+
(activation): SiLU(inplace=True)
|
868 |
+
(scale_activation): Sigmoid()
|
869 |
+
)
|
870 |
+
(3): Conv2dNormActivation(
|
871 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
872 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
873 |
+
)
|
874 |
+
)
|
875 |
+
(stochastic_depth): StochasticDepth(p=0.10379746835443039, mode=row)
|
876 |
+
)
|
877 |
+
(14): MBConv(
|
878 |
+
(block): Sequential(
|
879 |
+
(0): Conv2dNormActivation(
|
880 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
881 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
882 |
+
(2): SiLU(inplace=True)
|
883 |
+
)
|
884 |
+
(1): Conv2dNormActivation(
|
885 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
886 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
887 |
+
(2): SiLU(inplace=True)
|
888 |
+
)
|
889 |
+
(2): SqueezeExcitation(
|
890 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
891 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
892 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
893 |
+
(activation): SiLU(inplace=True)
|
894 |
+
(scale_activation): Sigmoid()
|
895 |
+
)
|
896 |
+
(3): Conv2dNormActivation(
|
897 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
898 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
899 |
+
)
|
900 |
+
)
|
901 |
+
(stochastic_depth): StochasticDepth(p=0.10632911392405063, mode=row)
|
902 |
+
)
|
903 |
+
(15): MBConv(
|
904 |
+
(block): Sequential(
|
905 |
+
(0): Conv2dNormActivation(
|
906 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
907 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
908 |
+
(2): SiLU(inplace=True)
|
909 |
+
)
|
910 |
+
(1): Conv2dNormActivation(
|
911 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
912 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
913 |
+
(2): SiLU(inplace=True)
|
914 |
+
)
|
915 |
+
(2): SqueezeExcitation(
|
916 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
917 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
918 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
919 |
+
(activation): SiLU(inplace=True)
|
920 |
+
(scale_activation): Sigmoid()
|
921 |
+
)
|
922 |
+
(3): Conv2dNormActivation(
|
923 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
924 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
925 |
+
)
|
926 |
+
)
|
927 |
+
(stochastic_depth): StochasticDepth(p=0.10886075949367088, mode=row)
|
928 |
+
)
|
929 |
+
(16): MBConv(
|
930 |
+
(block): Sequential(
|
931 |
+
(0): Conv2dNormActivation(
|
932 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
933 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
934 |
+
(2): SiLU(inplace=True)
|
935 |
+
)
|
936 |
+
(1): Conv2dNormActivation(
|
937 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
938 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
939 |
+
(2): SiLU(inplace=True)
|
940 |
+
)
|
941 |
+
(2): SqueezeExcitation(
|
942 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
943 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
944 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
945 |
+
(activation): SiLU(inplace=True)
|
946 |
+
(scale_activation): Sigmoid()
|
947 |
+
)
|
948 |
+
(3): Conv2dNormActivation(
|
949 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
950 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
951 |
+
)
|
952 |
+
)
|
953 |
+
(stochastic_depth): StochasticDepth(p=0.11139240506329115, mode=row)
|
954 |
+
)
|
955 |
+
(17): MBConv(
|
956 |
+
(block): Sequential(
|
957 |
+
(0): Conv2dNormActivation(
|
958 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
959 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
960 |
+
(2): SiLU(inplace=True)
|
961 |
+
)
|
962 |
+
(1): Conv2dNormActivation(
|
963 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
964 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
965 |
+
(2): SiLU(inplace=True)
|
966 |
+
)
|
967 |
+
(2): SqueezeExcitation(
|
968 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
969 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
970 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
971 |
+
(activation): SiLU(inplace=True)
|
972 |
+
(scale_activation): Sigmoid()
|
973 |
+
)
|
974 |
+
(3): Conv2dNormActivation(
|
975 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
976 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
977 |
+
)
|
978 |
+
)
|
979 |
+
(stochastic_depth): StochasticDepth(p=0.11392405063291139, mode=row)
|
980 |
+
)
|
981 |
+
(18): MBConv(
|
982 |
+
(block): Sequential(
|
983 |
+
(0): Conv2dNormActivation(
|
984 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
985 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
986 |
+
(2): SiLU(inplace=True)
|
987 |
+
)
|
988 |
+
(1): Conv2dNormActivation(
|
989 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1344, bias=False)
|
990 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
991 |
+
(2): SiLU(inplace=True)
|
992 |
+
)
|
993 |
+
(2): SqueezeExcitation(
|
994 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
995 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
996 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
997 |
+
(activation): SiLU(inplace=True)
|
998 |
+
(scale_activation): Sigmoid()
|
999 |
+
)
|
1000 |
+
(3): Conv2dNormActivation(
|
1001 |
+
(0): Conv2d(1344, 224, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1002 |
+
(1): BatchNorm2d(224, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1003 |
+
)
|
1004 |
+
)
|
1005 |
+
(stochastic_depth): StochasticDepth(p=0.11645569620253166, mode=row)
|
1006 |
+
)
|
1007 |
+
)
|
1008 |
+
(6): Sequential(
|
1009 |
+
(0): MBConv(
|
1010 |
+
(block): Sequential(
|
1011 |
+
(0): Conv2dNormActivation(
|
1012 |
+
(0): Conv2d(224, 1344, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1013 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1014 |
+
(2): SiLU(inplace=True)
|
1015 |
+
)
|
1016 |
+
(1): Conv2dNormActivation(
|
1017 |
+
(0): Conv2d(1344, 1344, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=1344, bias=False)
|
1018 |
+
(1): BatchNorm2d(1344, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1019 |
+
(2): SiLU(inplace=True)
|
1020 |
+
)
|
1021 |
+
(2): SqueezeExcitation(
|
1022 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1023 |
+
(fc1): Conv2d(1344, 56, kernel_size=(1, 1), stride=(1, 1))
|
1024 |
+
(fc2): Conv2d(56, 1344, kernel_size=(1, 1), stride=(1, 1))
|
1025 |
+
(activation): SiLU(inplace=True)
|
1026 |
+
(scale_activation): Sigmoid()
|
1027 |
+
)
|
1028 |
+
(3): Conv2dNormActivation(
|
1029 |
+
(0): Conv2d(1344, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1030 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1031 |
+
)
|
1032 |
+
)
|
1033 |
+
(stochastic_depth): StochasticDepth(p=0.11898734177215191, mode=row)
|
1034 |
+
)
|
1035 |
+
(1): MBConv(
|
1036 |
+
(block): Sequential(
|
1037 |
+
(0): Conv2dNormActivation(
|
1038 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1039 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1040 |
+
(2): SiLU(inplace=True)
|
1041 |
+
)
|
1042 |
+
(1): Conv2dNormActivation(
|
1043 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1044 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1045 |
+
(2): SiLU(inplace=True)
|
1046 |
+
)
|
1047 |
+
(2): SqueezeExcitation(
|
1048 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1049 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1050 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1051 |
+
(activation): SiLU(inplace=True)
|
1052 |
+
(scale_activation): Sigmoid()
|
1053 |
+
)
|
1054 |
+
(3): Conv2dNormActivation(
|
1055 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1056 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1057 |
+
)
|
1058 |
+
)
|
1059 |
+
(stochastic_depth): StochasticDepth(p=0.12151898734177217, mode=row)
|
1060 |
+
)
|
1061 |
+
(2): MBConv(
|
1062 |
+
(block): Sequential(
|
1063 |
+
(0): Conv2dNormActivation(
|
1064 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1065 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1066 |
+
(2): SiLU(inplace=True)
|
1067 |
+
)
|
1068 |
+
(1): Conv2dNormActivation(
|
1069 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1070 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1071 |
+
(2): SiLU(inplace=True)
|
1072 |
+
)
|
1073 |
+
(2): SqueezeExcitation(
|
1074 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1075 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1076 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1077 |
+
(activation): SiLU(inplace=True)
|
1078 |
+
(scale_activation): Sigmoid()
|
1079 |
+
)
|
1080 |
+
(3): Conv2dNormActivation(
|
1081 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1082 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1083 |
+
)
|
1084 |
+
)
|
1085 |
+
(stochastic_depth): StochasticDepth(p=0.12405063291139241, mode=row)
|
1086 |
+
)
|
1087 |
+
(3): MBConv(
|
1088 |
+
(block): Sequential(
|
1089 |
+
(0): Conv2dNormActivation(
|
1090 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1091 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1092 |
+
(2): SiLU(inplace=True)
|
1093 |
+
)
|
1094 |
+
(1): Conv2dNormActivation(
|
1095 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1096 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1097 |
+
(2): SiLU(inplace=True)
|
1098 |
+
)
|
1099 |
+
(2): SqueezeExcitation(
|
1100 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1101 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1102 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1103 |
+
(activation): SiLU(inplace=True)
|
1104 |
+
(scale_activation): Sigmoid()
|
1105 |
+
)
|
1106 |
+
(3): Conv2dNormActivation(
|
1107 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1108 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1109 |
+
)
|
1110 |
+
)
|
1111 |
+
(stochastic_depth): StochasticDepth(p=0.12658227848101267, mode=row)
|
1112 |
+
)
|
1113 |
+
(4): MBConv(
|
1114 |
+
(block): Sequential(
|
1115 |
+
(0): Conv2dNormActivation(
|
1116 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1117 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1118 |
+
(2): SiLU(inplace=True)
|
1119 |
+
)
|
1120 |
+
(1): Conv2dNormActivation(
|
1121 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1122 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1123 |
+
(2): SiLU(inplace=True)
|
1124 |
+
)
|
1125 |
+
(2): SqueezeExcitation(
|
1126 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1127 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1128 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1129 |
+
(activation): SiLU(inplace=True)
|
1130 |
+
(scale_activation): Sigmoid()
|
1131 |
+
)
|
1132 |
+
(3): Conv2dNormActivation(
|
1133 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1134 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1135 |
+
)
|
1136 |
+
)
|
1137 |
+
(stochastic_depth): StochasticDepth(p=0.12911392405063293, mode=row)
|
1138 |
+
)
|
1139 |
+
(5): MBConv(
|
1140 |
+
(block): Sequential(
|
1141 |
+
(0): Conv2dNormActivation(
|
1142 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1143 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1144 |
+
(2): SiLU(inplace=True)
|
1145 |
+
)
|
1146 |
+
(1): Conv2dNormActivation(
|
1147 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1148 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1149 |
+
(2): SiLU(inplace=True)
|
1150 |
+
)
|
1151 |
+
(2): SqueezeExcitation(
|
1152 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1153 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1154 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1155 |
+
(activation): SiLU(inplace=True)
|
1156 |
+
(scale_activation): Sigmoid()
|
1157 |
+
)
|
1158 |
+
(3): Conv2dNormActivation(
|
1159 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1160 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1161 |
+
)
|
1162 |
+
)
|
1163 |
+
(stochastic_depth): StochasticDepth(p=0.13164556962025317, mode=row)
|
1164 |
+
)
|
1165 |
+
(6): MBConv(
|
1166 |
+
(block): Sequential(
|
1167 |
+
(0): Conv2dNormActivation(
|
1168 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1169 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1170 |
+
(2): SiLU(inplace=True)
|
1171 |
+
)
|
1172 |
+
(1): Conv2dNormActivation(
|
1173 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1174 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1175 |
+
(2): SiLU(inplace=True)
|
1176 |
+
)
|
1177 |
+
(2): SqueezeExcitation(
|
1178 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1179 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1180 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1181 |
+
(activation): SiLU(inplace=True)
|
1182 |
+
(scale_activation): Sigmoid()
|
1183 |
+
)
|
1184 |
+
(3): Conv2dNormActivation(
|
1185 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1186 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1187 |
+
)
|
1188 |
+
)
|
1189 |
+
(stochastic_depth): StochasticDepth(p=0.13417721518987344, mode=row)
|
1190 |
+
)
|
1191 |
+
(7): MBConv(
|
1192 |
+
(block): Sequential(
|
1193 |
+
(0): Conv2dNormActivation(
|
1194 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1195 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1196 |
+
(2): SiLU(inplace=True)
|
1197 |
+
)
|
1198 |
+
(1): Conv2dNormActivation(
|
1199 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1200 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1201 |
+
(2): SiLU(inplace=True)
|
1202 |
+
)
|
1203 |
+
(2): SqueezeExcitation(
|
1204 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1205 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1206 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1207 |
+
(activation): SiLU(inplace=True)
|
1208 |
+
(scale_activation): Sigmoid()
|
1209 |
+
)
|
1210 |
+
(3): Conv2dNormActivation(
|
1211 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1212 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1213 |
+
)
|
1214 |
+
)
|
1215 |
+
(stochastic_depth): StochasticDepth(p=0.13670886075949368, mode=row)
|
1216 |
+
)
|
1217 |
+
(8): MBConv(
|
1218 |
+
(block): Sequential(
|
1219 |
+
(0): Conv2dNormActivation(
|
1220 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1221 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1222 |
+
(2): SiLU(inplace=True)
|
1223 |
+
)
|
1224 |
+
(1): Conv2dNormActivation(
|
1225 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1226 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1227 |
+
(2): SiLU(inplace=True)
|
1228 |
+
)
|
1229 |
+
(2): SqueezeExcitation(
|
1230 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1231 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1232 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1233 |
+
(activation): SiLU(inplace=True)
|
1234 |
+
(scale_activation): Sigmoid()
|
1235 |
+
)
|
1236 |
+
(3): Conv2dNormActivation(
|
1237 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1238 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1239 |
+
)
|
1240 |
+
)
|
1241 |
+
(stochastic_depth): StochasticDepth(p=0.13924050632911392, mode=row)
|
1242 |
+
)
|
1243 |
+
(9): MBConv(
|
1244 |
+
(block): Sequential(
|
1245 |
+
(0): Conv2dNormActivation(
|
1246 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1247 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1248 |
+
(2): SiLU(inplace=True)
|
1249 |
+
)
|
1250 |
+
(1): Conv2dNormActivation(
|
1251 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1252 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1253 |
+
(2): SiLU(inplace=True)
|
1254 |
+
)
|
1255 |
+
(2): SqueezeExcitation(
|
1256 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1257 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1258 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1259 |
+
(activation): SiLU(inplace=True)
|
1260 |
+
(scale_activation): Sigmoid()
|
1261 |
+
)
|
1262 |
+
(3): Conv2dNormActivation(
|
1263 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1264 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1265 |
+
)
|
1266 |
+
)
|
1267 |
+
(stochastic_depth): StochasticDepth(p=0.14177215189873418, mode=row)
|
1268 |
+
)
|
1269 |
+
(10): MBConv(
|
1270 |
+
(block): Sequential(
|
1271 |
+
(0): Conv2dNormActivation(
|
1272 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1273 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1274 |
+
(2): SiLU(inplace=True)
|
1275 |
+
)
|
1276 |
+
(1): Conv2dNormActivation(
|
1277 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1278 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1279 |
+
(2): SiLU(inplace=True)
|
1280 |
+
)
|
1281 |
+
(2): SqueezeExcitation(
|
1282 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1283 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1284 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1285 |
+
(activation): SiLU(inplace=True)
|
1286 |
+
(scale_activation): Sigmoid()
|
1287 |
+
)
|
1288 |
+
(3): Conv2dNormActivation(
|
1289 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1290 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1291 |
+
)
|
1292 |
+
)
|
1293 |
+
(stochastic_depth): StochasticDepth(p=0.14430379746835442, mode=row)
|
1294 |
+
)
|
1295 |
+
(11): MBConv(
|
1296 |
+
(block): Sequential(
|
1297 |
+
(0): Conv2dNormActivation(
|
1298 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1299 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1300 |
+
(2): SiLU(inplace=True)
|
1301 |
+
)
|
1302 |
+
(1): Conv2dNormActivation(
|
1303 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1304 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1305 |
+
(2): SiLU(inplace=True)
|
1306 |
+
)
|
1307 |
+
(2): SqueezeExcitation(
|
1308 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1309 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1310 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1311 |
+
(activation): SiLU(inplace=True)
|
1312 |
+
(scale_activation): Sigmoid()
|
1313 |
+
)
|
1314 |
+
(3): Conv2dNormActivation(
|
1315 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1316 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1317 |
+
)
|
1318 |
+
)
|
1319 |
+
(stochastic_depth): StochasticDepth(p=0.1468354430379747, mode=row)
|
1320 |
+
)
|
1321 |
+
(12): MBConv(
|
1322 |
+
(block): Sequential(
|
1323 |
+
(0): Conv2dNormActivation(
|
1324 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1325 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1326 |
+
(2): SiLU(inplace=True)
|
1327 |
+
)
|
1328 |
+
(1): Conv2dNormActivation(
|
1329 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1330 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1331 |
+
(2): SiLU(inplace=True)
|
1332 |
+
)
|
1333 |
+
(2): SqueezeExcitation(
|
1334 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1335 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1336 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1337 |
+
(activation): SiLU(inplace=True)
|
1338 |
+
(scale_activation): Sigmoid()
|
1339 |
+
)
|
1340 |
+
(3): Conv2dNormActivation(
|
1341 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1342 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1343 |
+
)
|
1344 |
+
)
|
1345 |
+
(stochastic_depth): StochasticDepth(p=0.14936708860759496, mode=row)
|
1346 |
+
)
|
1347 |
+
(13): MBConv(
|
1348 |
+
(block): Sequential(
|
1349 |
+
(0): Conv2dNormActivation(
|
1350 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1351 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1352 |
+
(2): SiLU(inplace=True)
|
1353 |
+
)
|
1354 |
+
(1): Conv2dNormActivation(
|
1355 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1356 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1357 |
+
(2): SiLU(inplace=True)
|
1358 |
+
)
|
1359 |
+
(2): SqueezeExcitation(
|
1360 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1361 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1362 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1363 |
+
(activation): SiLU(inplace=True)
|
1364 |
+
(scale_activation): Sigmoid()
|
1365 |
+
)
|
1366 |
+
(3): Conv2dNormActivation(
|
1367 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1368 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1369 |
+
)
|
1370 |
+
)
|
1371 |
+
(stochastic_depth): StochasticDepth(p=0.1518987341772152, mode=row)
|
1372 |
+
)
|
1373 |
+
(14): MBConv(
|
1374 |
+
(block): Sequential(
|
1375 |
+
(0): Conv2dNormActivation(
|
1376 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1377 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1378 |
+
(2): SiLU(inplace=True)
|
1379 |
+
)
|
1380 |
+
(1): Conv2dNormActivation(
|
1381 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1382 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1383 |
+
(2): SiLU(inplace=True)
|
1384 |
+
)
|
1385 |
+
(2): SqueezeExcitation(
|
1386 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1387 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1388 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1389 |
+
(activation): SiLU(inplace=True)
|
1390 |
+
(scale_activation): Sigmoid()
|
1391 |
+
)
|
1392 |
+
(3): Conv2dNormActivation(
|
1393 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1394 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1395 |
+
)
|
1396 |
+
)
|
1397 |
+
(stochastic_depth): StochasticDepth(p=0.15443037974683546, mode=row)
|
1398 |
+
)
|
1399 |
+
(15): MBConv(
|
1400 |
+
(block): Sequential(
|
1401 |
+
(0): Conv2dNormActivation(
|
1402 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1403 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1404 |
+
(2): SiLU(inplace=True)
|
1405 |
+
)
|
1406 |
+
(1): Conv2dNormActivation(
|
1407 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1408 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1409 |
+
(2): SiLU(inplace=True)
|
1410 |
+
)
|
1411 |
+
(2): SqueezeExcitation(
|
1412 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1413 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1414 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1415 |
+
(activation): SiLU(inplace=True)
|
1416 |
+
(scale_activation): Sigmoid()
|
1417 |
+
)
|
1418 |
+
(3): Conv2dNormActivation(
|
1419 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1420 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1421 |
+
)
|
1422 |
+
)
|
1423 |
+
(stochastic_depth): StochasticDepth(p=0.1569620253164557, mode=row)
|
1424 |
+
)
|
1425 |
+
(16): MBConv(
|
1426 |
+
(block): Sequential(
|
1427 |
+
(0): Conv2dNormActivation(
|
1428 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1429 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1430 |
+
(2): SiLU(inplace=True)
|
1431 |
+
)
|
1432 |
+
(1): Conv2dNormActivation(
|
1433 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1434 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1435 |
+
(2): SiLU(inplace=True)
|
1436 |
+
)
|
1437 |
+
(2): SqueezeExcitation(
|
1438 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1439 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1440 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1441 |
+
(activation): SiLU(inplace=True)
|
1442 |
+
(scale_activation): Sigmoid()
|
1443 |
+
)
|
1444 |
+
(3): Conv2dNormActivation(
|
1445 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1446 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1447 |
+
)
|
1448 |
+
)
|
1449 |
+
(stochastic_depth): StochasticDepth(p=0.15949367088607597, mode=row)
|
1450 |
+
)
|
1451 |
+
(17): MBConv(
|
1452 |
+
(block): Sequential(
|
1453 |
+
(0): Conv2dNormActivation(
|
1454 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1455 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1456 |
+
(2): SiLU(inplace=True)
|
1457 |
+
)
|
1458 |
+
(1): Conv2dNormActivation(
|
1459 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1460 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1461 |
+
(2): SiLU(inplace=True)
|
1462 |
+
)
|
1463 |
+
(2): SqueezeExcitation(
|
1464 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1465 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1466 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1467 |
+
(activation): SiLU(inplace=True)
|
1468 |
+
(scale_activation): Sigmoid()
|
1469 |
+
)
|
1470 |
+
(3): Conv2dNormActivation(
|
1471 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1472 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1473 |
+
)
|
1474 |
+
)
|
1475 |
+
(stochastic_depth): StochasticDepth(p=0.1620253164556962, mode=row)
|
1476 |
+
)
|
1477 |
+
(18): MBConv(
|
1478 |
+
(block): Sequential(
|
1479 |
+
(0): Conv2dNormActivation(
|
1480 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1481 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1482 |
+
(2): SiLU(inplace=True)
|
1483 |
+
)
|
1484 |
+
(1): Conv2dNormActivation(
|
1485 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1486 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1487 |
+
(2): SiLU(inplace=True)
|
1488 |
+
)
|
1489 |
+
(2): SqueezeExcitation(
|
1490 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1491 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1492 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1493 |
+
(activation): SiLU(inplace=True)
|
1494 |
+
(scale_activation): Sigmoid()
|
1495 |
+
)
|
1496 |
+
(3): Conv2dNormActivation(
|
1497 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1498 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1499 |
+
)
|
1500 |
+
)
|
1501 |
+
(stochastic_depth): StochasticDepth(p=0.16455696202531644, mode=row)
|
1502 |
+
)
|
1503 |
+
(19): MBConv(
|
1504 |
+
(block): Sequential(
|
1505 |
+
(0): Conv2dNormActivation(
|
1506 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1507 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1508 |
+
(2): SiLU(inplace=True)
|
1509 |
+
)
|
1510 |
+
(1): Conv2dNormActivation(
|
1511 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1512 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1513 |
+
(2): SiLU(inplace=True)
|
1514 |
+
)
|
1515 |
+
(2): SqueezeExcitation(
|
1516 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1517 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1518 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1519 |
+
(activation): SiLU(inplace=True)
|
1520 |
+
(scale_activation): Sigmoid()
|
1521 |
+
)
|
1522 |
+
(3): Conv2dNormActivation(
|
1523 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1524 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1525 |
+
)
|
1526 |
+
)
|
1527 |
+
(stochastic_depth): StochasticDepth(p=0.1670886075949367, mode=row)
|
1528 |
+
)
|
1529 |
+
(20): MBConv(
|
1530 |
+
(block): Sequential(
|
1531 |
+
(0): Conv2dNormActivation(
|
1532 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1533 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1534 |
+
(2): SiLU(inplace=True)
|
1535 |
+
)
|
1536 |
+
(1): Conv2dNormActivation(
|
1537 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1538 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1539 |
+
(2): SiLU(inplace=True)
|
1540 |
+
)
|
1541 |
+
(2): SqueezeExcitation(
|
1542 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1543 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1544 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1545 |
+
(activation): SiLU(inplace=True)
|
1546 |
+
(scale_activation): Sigmoid()
|
1547 |
+
)
|
1548 |
+
(3): Conv2dNormActivation(
|
1549 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1550 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1551 |
+
)
|
1552 |
+
)
|
1553 |
+
(stochastic_depth): StochasticDepth(p=0.16962025316455698, mode=row)
|
1554 |
+
)
|
1555 |
+
(21): MBConv(
|
1556 |
+
(block): Sequential(
|
1557 |
+
(0): Conv2dNormActivation(
|
1558 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1559 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1560 |
+
(2): SiLU(inplace=True)
|
1561 |
+
)
|
1562 |
+
(1): Conv2dNormActivation(
|
1563 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1564 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1565 |
+
(2): SiLU(inplace=True)
|
1566 |
+
)
|
1567 |
+
(2): SqueezeExcitation(
|
1568 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1569 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1570 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1571 |
+
(activation): SiLU(inplace=True)
|
1572 |
+
(scale_activation): Sigmoid()
|
1573 |
+
)
|
1574 |
+
(3): Conv2dNormActivation(
|
1575 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1576 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1577 |
+
)
|
1578 |
+
)
|
1579 |
+
(stochastic_depth): StochasticDepth(p=0.17215189873417724, mode=row)
|
1580 |
+
)
|
1581 |
+
(22): MBConv(
|
1582 |
+
(block): Sequential(
|
1583 |
+
(0): Conv2dNormActivation(
|
1584 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1585 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1586 |
+
(2): SiLU(inplace=True)
|
1587 |
+
)
|
1588 |
+
(1): Conv2dNormActivation(
|
1589 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1590 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1591 |
+
(2): SiLU(inplace=True)
|
1592 |
+
)
|
1593 |
+
(2): SqueezeExcitation(
|
1594 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1595 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1596 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1597 |
+
(activation): SiLU(inplace=True)
|
1598 |
+
(scale_activation): Sigmoid()
|
1599 |
+
)
|
1600 |
+
(3): Conv2dNormActivation(
|
1601 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1602 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1603 |
+
)
|
1604 |
+
)
|
1605 |
+
(stochastic_depth): StochasticDepth(p=0.17468354430379748, mode=row)
|
1606 |
+
)
|
1607 |
+
(23): MBConv(
|
1608 |
+
(block): Sequential(
|
1609 |
+
(0): Conv2dNormActivation(
|
1610 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1611 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1612 |
+
(2): SiLU(inplace=True)
|
1613 |
+
)
|
1614 |
+
(1): Conv2dNormActivation(
|
1615 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1616 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1617 |
+
(2): SiLU(inplace=True)
|
1618 |
+
)
|
1619 |
+
(2): SqueezeExcitation(
|
1620 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1621 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1622 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1623 |
+
(activation): SiLU(inplace=True)
|
1624 |
+
(scale_activation): Sigmoid()
|
1625 |
+
)
|
1626 |
+
(3): Conv2dNormActivation(
|
1627 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1628 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1629 |
+
)
|
1630 |
+
)
|
1631 |
+
(stochastic_depth): StochasticDepth(p=0.17721518987341772, mode=row)
|
1632 |
+
)
|
1633 |
+
(24): MBConv(
|
1634 |
+
(block): Sequential(
|
1635 |
+
(0): Conv2dNormActivation(
|
1636 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1637 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1638 |
+
(2): SiLU(inplace=True)
|
1639 |
+
)
|
1640 |
+
(1): Conv2dNormActivation(
|
1641 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1642 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1643 |
+
(2): SiLU(inplace=True)
|
1644 |
+
)
|
1645 |
+
(2): SqueezeExcitation(
|
1646 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1647 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1648 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1649 |
+
(activation): SiLU(inplace=True)
|
1650 |
+
(scale_activation): Sigmoid()
|
1651 |
+
)
|
1652 |
+
(3): Conv2dNormActivation(
|
1653 |
+
(0): Conv2d(2304, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1654 |
+
(1): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1655 |
+
)
|
1656 |
+
)
|
1657 |
+
(stochastic_depth): StochasticDepth(p=0.179746835443038, mode=row)
|
1658 |
+
)
|
1659 |
+
)
|
1660 |
+
(7): Sequential(
|
1661 |
+
(0): MBConv(
|
1662 |
+
(block): Sequential(
|
1663 |
+
(0): Conv2dNormActivation(
|
1664 |
+
(0): Conv2d(384, 2304, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1665 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1666 |
+
(2): SiLU(inplace=True)
|
1667 |
+
)
|
1668 |
+
(1): Conv2dNormActivation(
|
1669 |
+
(0): Conv2d(2304, 2304, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2304, bias=False)
|
1670 |
+
(1): BatchNorm2d(2304, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1671 |
+
(2): SiLU(inplace=True)
|
1672 |
+
)
|
1673 |
+
(2): SqueezeExcitation(
|
1674 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1675 |
+
(fc1): Conv2d(2304, 96, kernel_size=(1, 1), stride=(1, 1))
|
1676 |
+
(fc2): Conv2d(96, 2304, kernel_size=(1, 1), stride=(1, 1))
|
1677 |
+
(activation): SiLU(inplace=True)
|
1678 |
+
(scale_activation): Sigmoid()
|
1679 |
+
)
|
1680 |
+
(3): Conv2dNormActivation(
|
1681 |
+
(0): Conv2d(2304, 640, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1682 |
+
(1): BatchNorm2d(640, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1683 |
+
)
|
1684 |
+
)
|
1685 |
+
(stochastic_depth): StochasticDepth(p=0.18227848101265823, mode=row)
|
1686 |
+
)
|
1687 |
+
(1): MBConv(
|
1688 |
+
(block): Sequential(
|
1689 |
+
(0): Conv2dNormActivation(
|
1690 |
+
(0): Conv2d(640, 3840, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1691 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1692 |
+
(2): SiLU(inplace=True)
|
1693 |
+
)
|
1694 |
+
(1): Conv2dNormActivation(
|
1695 |
+
(0): Conv2d(3840, 3840, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=3840, bias=False)
|
1696 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1697 |
+
(2): SiLU(inplace=True)
|
1698 |
+
)
|
1699 |
+
(2): SqueezeExcitation(
|
1700 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1701 |
+
(fc1): Conv2d(3840, 160, kernel_size=(1, 1), stride=(1, 1))
|
1702 |
+
(fc2): Conv2d(160, 3840, kernel_size=(1, 1), stride=(1, 1))
|
1703 |
+
(activation): SiLU(inplace=True)
|
1704 |
+
(scale_activation): Sigmoid()
|
1705 |
+
)
|
1706 |
+
(3): Conv2dNormActivation(
|
1707 |
+
(0): Conv2d(3840, 640, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1708 |
+
(1): BatchNorm2d(640, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1709 |
+
)
|
1710 |
+
)
|
1711 |
+
(stochastic_depth): StochasticDepth(p=0.1848101265822785, mode=row)
|
1712 |
+
)
|
1713 |
+
(2): MBConv(
|
1714 |
+
(block): Sequential(
|
1715 |
+
(0): Conv2dNormActivation(
|
1716 |
+
(0): Conv2d(640, 3840, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1717 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1718 |
+
(2): SiLU(inplace=True)
|
1719 |
+
)
|
1720 |
+
(1): Conv2dNormActivation(
|
1721 |
+
(0): Conv2d(3840, 3840, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=3840, bias=False)
|
1722 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1723 |
+
(2): SiLU(inplace=True)
|
1724 |
+
)
|
1725 |
+
(2): SqueezeExcitation(
|
1726 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1727 |
+
(fc1): Conv2d(3840, 160, kernel_size=(1, 1), stride=(1, 1))
|
1728 |
+
(fc2): Conv2d(160, 3840, kernel_size=(1, 1), stride=(1, 1))
|
1729 |
+
(activation): SiLU(inplace=True)
|
1730 |
+
(scale_activation): Sigmoid()
|
1731 |
+
)
|
1732 |
+
(3): Conv2dNormActivation(
|
1733 |
+
(0): Conv2d(3840, 640, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1734 |
+
(1): BatchNorm2d(640, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1735 |
+
)
|
1736 |
+
)
|
1737 |
+
(stochastic_depth): StochasticDepth(p=0.18734177215189873, mode=row)
|
1738 |
+
)
|
1739 |
+
(3): MBConv(
|
1740 |
+
(block): Sequential(
|
1741 |
+
(0): Conv2dNormActivation(
|
1742 |
+
(0): Conv2d(640, 3840, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1743 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1744 |
+
(2): SiLU(inplace=True)
|
1745 |
+
)
|
1746 |
+
(1): Conv2dNormActivation(
|
1747 |
+
(0): Conv2d(3840, 3840, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=3840, bias=False)
|
1748 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1749 |
+
(2): SiLU(inplace=True)
|
1750 |
+
)
|
1751 |
+
(2): SqueezeExcitation(
|
1752 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1753 |
+
(fc1): Conv2d(3840, 160, kernel_size=(1, 1), stride=(1, 1))
|
1754 |
+
(fc2): Conv2d(160, 3840, kernel_size=(1, 1), stride=(1, 1))
|
1755 |
+
(activation): SiLU(inplace=True)
|
1756 |
+
(scale_activation): Sigmoid()
|
1757 |
+
)
|
1758 |
+
(3): Conv2dNormActivation(
|
1759 |
+
(0): Conv2d(3840, 640, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1760 |
+
(1): BatchNorm2d(640, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1761 |
+
)
|
1762 |
+
)
|
1763 |
+
(stochastic_depth): StochasticDepth(p=0.189873417721519, mode=row)
|
1764 |
+
)
|
1765 |
+
(4): MBConv(
|
1766 |
+
(block): Sequential(
|
1767 |
+
(0): Conv2dNormActivation(
|
1768 |
+
(0): Conv2d(640, 3840, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1769 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1770 |
+
(2): SiLU(inplace=True)
|
1771 |
+
)
|
1772 |
+
(1): Conv2dNormActivation(
|
1773 |
+
(0): Conv2d(3840, 3840, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=3840, bias=False)
|
1774 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1775 |
+
(2): SiLU(inplace=True)
|
1776 |
+
)
|
1777 |
+
(2): SqueezeExcitation(
|
1778 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1779 |
+
(fc1): Conv2d(3840, 160, kernel_size=(1, 1), stride=(1, 1))
|
1780 |
+
(fc2): Conv2d(160, 3840, kernel_size=(1, 1), stride=(1, 1))
|
1781 |
+
(activation): SiLU(inplace=True)
|
1782 |
+
(scale_activation): Sigmoid()
|
1783 |
+
)
|
1784 |
+
(3): Conv2dNormActivation(
|
1785 |
+
(0): Conv2d(3840, 640, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1786 |
+
(1): BatchNorm2d(640, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1787 |
+
)
|
1788 |
+
)
|
1789 |
+
(stochastic_depth): StochasticDepth(p=0.19240506329113927, mode=row)
|
1790 |
+
)
|
1791 |
+
(5): MBConv(
|
1792 |
+
(block): Sequential(
|
1793 |
+
(0): Conv2dNormActivation(
|
1794 |
+
(0): Conv2d(640, 3840, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1795 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1796 |
+
(2): SiLU(inplace=True)
|
1797 |
+
)
|
1798 |
+
(1): Conv2dNormActivation(
|
1799 |
+
(0): Conv2d(3840, 3840, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=3840, bias=False)
|
1800 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1801 |
+
(2): SiLU(inplace=True)
|
1802 |
+
)
|
1803 |
+
(2): SqueezeExcitation(
|
1804 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1805 |
+
(fc1): Conv2d(3840, 160, kernel_size=(1, 1), stride=(1, 1))
|
1806 |
+
(fc2): Conv2d(160, 3840, kernel_size=(1, 1), stride=(1, 1))
|
1807 |
+
(activation): SiLU(inplace=True)
|
1808 |
+
(scale_activation): Sigmoid()
|
1809 |
+
)
|
1810 |
+
(3): Conv2dNormActivation(
|
1811 |
+
(0): Conv2d(3840, 640, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1812 |
+
(1): BatchNorm2d(640, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1813 |
+
)
|
1814 |
+
)
|
1815 |
+
(stochastic_depth): StochasticDepth(p=0.1949367088607595, mode=row)
|
1816 |
+
)
|
1817 |
+
(6): MBConv(
|
1818 |
+
(block): Sequential(
|
1819 |
+
(0): Conv2dNormActivation(
|
1820 |
+
(0): Conv2d(640, 3840, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1821 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1822 |
+
(2): SiLU(inplace=True)
|
1823 |
+
)
|
1824 |
+
(1): Conv2dNormActivation(
|
1825 |
+
(0): Conv2d(3840, 3840, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=3840, bias=False)
|
1826 |
+
(1): BatchNorm2d(3840, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1827 |
+
(2): SiLU(inplace=True)
|
1828 |
+
)
|
1829 |
+
(2): SqueezeExcitation(
|
1830 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1831 |
+
(fc1): Conv2d(3840, 160, kernel_size=(1, 1), stride=(1, 1))
|
1832 |
+
(fc2): Conv2d(160, 3840, kernel_size=(1, 1), stride=(1, 1))
|
1833 |
+
(activation): SiLU(inplace=True)
|
1834 |
+
(scale_activation): Sigmoid()
|
1835 |
+
)
|
1836 |
+
(3): Conv2dNormActivation(
|
1837 |
+
(0): Conv2d(3840, 640, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1838 |
+
(1): BatchNorm2d(640, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1839 |
+
)
|
1840 |
+
)
|
1841 |
+
(stochastic_depth): StochasticDepth(p=0.19746835443037977, mode=row)
|
1842 |
+
)
|
1843 |
+
)
|
1844 |
+
(8): Conv2dNormActivation(
|
1845 |
+
(0): Conv2d(640, 1280, kernel_size=(1, 1), stride=(1, 1), bias=False)
|
1846 |
+
(1): BatchNorm2d(1280, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
|
1847 |
+
(2): SiLU(inplace=True)
|
1848 |
+
)
|
1849 |
+
)
|
1850 |
+
(avgpool): AdaptiveAvgPool2d(output_size=1)
|
1851 |
+
(classifier): Sequential(
|
1852 |
+
(0): Dropout(p=0.4, inplace=True)
|
1853 |
+
(1): Linear(in_features=1280, out_features=25, bias=True)
|
1854 |
+
)
|
1855 |
+
)
|
requirements.txt
ADDED
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
aiohttp==3.8.1
|
2 |
+
aiosignal==1.2.0
|
3 |
+
analytics-python==1.4.0
|
4 |
+
anyio==3.6.1
|
5 |
+
async-timeout==4.0.2
|
6 |
+
attrs==22.1.0
|
7 |
+
autopep8==1.6.0
|
8 |
+
backoff==1.10.0
|
9 |
+
bcrypt==3.2.2
|
10 |
+
certifi==2022.6.15
|
11 |
+
cffi==1.15.1
|
12 |
+
charset-normalizer==2.1.0
|
13 |
+
click==8.1.3
|
14 |
+
colorama==0.4.5
|
15 |
+
cryptography==37.0.4
|
16 |
+
cycler==0.11.0
|
17 |
+
fastapi==0.79.0
|
18 |
+
ffmpy==0.3.0
|
19 |
+
fonttools==4.34.4
|
20 |
+
frozenlist==1.3.1
|
21 |
+
fsspec==2022.7.1
|
22 |
+
grad-cam==1.4.2
|
23 |
+
gradio==3.1.4
|
24 |
+
h11==0.12.0
|
25 |
+
httpcore==0.15.0
|
26 |
+
httpx==0.23.0
|
27 |
+
idna==3.3
|
28 |
+
Jinja2==3.1.2
|
29 |
+
joblib==1.1.0
|
30 |
+
kiwisolver==1.4.4
|
31 |
+
linkify-it-py==1.0.3
|
32 |
+
markdown-it-py==2.1.0
|
33 |
+
MarkupSafe==2.1.1
|
34 |
+
matplotlib==3.5.2
|
35 |
+
mdit-py-plugins==0.3.0
|
36 |
+
mdurl==0.1.1
|
37 |
+
monotonic==1.6
|
38 |
+
multidict==6.0.2
|
39 |
+
numpy==1.23.1
|
40 |
+
opencv-python==4.6.0.66
|
41 |
+
orjson==3.7.11
|
42 |
+
packaging==21.3
|
43 |
+
pandas==1.4.3
|
44 |
+
paramiko==2.11.0
|
45 |
+
Pillow==9.2.0
|
46 |
+
pycodestyle==2.9.1
|
47 |
+
pycparser==2.21
|
48 |
+
pycryptodome==3.15.0
|
49 |
+
pydantic==1.9.1
|
50 |
+
pydub==0.25.1
|
51 |
+
PyNaCl==1.5.0
|
52 |
+
pyparsing==3.0.9
|
53 |
+
python-dateutil==2.8.2
|
54 |
+
python-multipart==0.0.5
|
55 |
+
pytz==2022.1
|
56 |
+
requests==2.28.1
|
57 |
+
rfc3986==1.5.0
|
58 |
+
scikit-learn==1.1.2
|
59 |
+
scipy==1.9.0
|
60 |
+
six==1.16.0
|
61 |
+
sniffio==1.2.0
|
62 |
+
starlette==0.19.1
|
63 |
+
threadpoolctl==3.1.0
|
64 |
+
toml==0.10.2
|
65 |
+
torch==1.12.1
|
66 |
+
torchaudio==0.12.1
|
67 |
+
torchvision==0.13.1
|
68 |
+
tqdm==4.64.0
|
69 |
+
ttach==0.0.3
|
70 |
+
typing_extensions==4.3.0
|
71 |
+
uc-micro-py==1.0.1
|
72 |
+
urllib3==1.26.11
|
73 |
+
uvicorn==0.18.2
|
74 |
+
yarl==1.8.1
|
utils/__init__.py
ADDED
File without changes
|
utils/__pycache__/__init__.cpython-310.pyc
ADDED
Binary file (148 Bytes). View file
|
utils/__pycache__/imshow.cpython-310.pyc
ADDED
Binary file (752 Bytes). View file
|
utils/__pycache__/save_load.cpython-310.pyc
ADDED
Binary file (549 Bytes). View file
|
utils/__pycache__/utils.cpython-310.pyc
ADDED
Binary file (411 Bytes). View file
|
utils/imshow.py
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from matplotlib import pyplot as plt
|
2 |
+
import numpy as np
|
3 |
+
import torchvision
|
4 |
+
|
5 |
+
|
6 |
+
def imshow(dataloader, title=None):
|
7 |
+
inputs, _ = next(iter(dataloader))
|
8 |
+
out = torchvision.utils.make_grid(inputs)
|
9 |
+
inp = out.numpy().transpose((1, 2, 0))
|
10 |
+
mean = np.array([0.485, 0.456, 0.406])
|
11 |
+
std = np.array([0.229, 0.224, 0.225])
|
12 |
+
inp = std * inp + mean
|
13 |
+
inp = np.clip(inp, 0, 1)
|
14 |
+
plt.imshow(inp)
|
15 |
+
if title is not None:
|
16 |
+
plt.title(title)
|
17 |
+
plt.show()
|
18 |
+
plt.pause(0.001) # pause a bit so that plots are updated
|
utils/save_load.py
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
import torch
|
3 |
+
|
4 |
+
|
5 |
+
def save_model(model):
|
6 |
+
torch.save(model.state_dict(), 'model_weights.pth')
|
7 |
+
|
8 |
+
|
9 |
+
def load_model(model):
|
10 |
+
return model.load_state_dict(torch.load('./models/model_weights_27_styles.pth', map_location=torch.device('cpu')))
|