Spaces:
Sleeping
Sleeping
balajib197
commited on
Commit
•
2581e2d
1
Parent(s):
e8111f3
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch, torchvision
|
2 |
+
from torchvision import transforms
|
3 |
+
import numpy as np
|
4 |
+
import gradio as gr
|
5 |
+
from PIL import Image
|
6 |
+
from pytorch_grad_cam import GradCAM
|
7 |
+
from pytorch_grad_cam.utils.image import show_cam_on_image
|
8 |
+
from resnet import LitResnet, BasicBlock
|
9 |
+
import gradio as gr
|
10 |
+
import matplotlib.pyplot as plt
|
11 |
+
|
12 |
+
model = LitResnet(BasicBlock, [2, 2, 2, 2], lr=0.05)
|
13 |
+
model = LitResnet.load_from_checkpoint("model_60.ckpt")
|
14 |
+
|
15 |
+
inv_normalize = transforms.Normalize(
|
16 |
+
mean=[-0.50/0.23, -0.50/0.23, -0.50/0.23],
|
17 |
+
std=[1/0.23, 1/0.23, 1/0.23]
|
18 |
+
)
|
19 |
+
classes = ('plane', 'car', 'bird', 'cat', 'deer',
|
20 |
+
'dog', 'frog', 'horse', 'ship', 'truck')
|
21 |
+
|
22 |
+
def resize_image_pil(image, new_width, new_height):
|
23 |
+
|
24 |
+
# Convert to PIL image
|
25 |
+
img = Image.fromarray(np.array(image))
|
26 |
+
|
27 |
+
# Get original size
|
28 |
+
width, height = img.size
|
29 |
+
|
30 |
+
# Calculate scale
|
31 |
+
width_scale = new_width / width
|
32 |
+
height_scale = new_height / height
|
33 |
+
scale = min(width_scale, height_scale)
|
34 |
+
|
35 |
+
# Resize
|
36 |
+
resized = img.resize((int(width*scale), int(height*scale)), Image.NEAREST)
|
37 |
+
|
38 |
+
# Crop to exact size
|
39 |
+
resized = resized.crop((0, 0, new_width, new_height))
|
40 |
+
|
41 |
+
return resized
|
42 |
+
|
43 |
+
def inference(input_img, gradcam_res, top_res, transparency = 0.5, target_layer_number = -1):
|
44 |
+
input_img = resize_image_pil(input_img, 32, 32)
|
45 |
+
|
46 |
+
input_img = np.array(input_img)
|
47 |
+
org_img = input_img
|
48 |
+
input_img = input_img.reshape((32, 32, 3))
|
49 |
+
transform = transforms.ToTensor()
|
50 |
+
input_img = transform(input_img)
|
51 |
+
input_img = input_img
|
52 |
+
input_img = input_img.unsqueeze(0)
|
53 |
+
outputs = model(input_img)
|
54 |
+
softmax = torch.nn.Softmax(dim=0)
|
55 |
+
o = softmax(outputs.flatten())
|
56 |
+
confidences = {classes[i]: float(o[i]) for i in range(10)}
|
57 |
+
_, prediction = torch.max(outputs, 1)
|
58 |
+
target_layers = [model.layer2[target_layer_number]]
|
59 |
+
cam = GradCAM(model=model, target_layers=target_layers)
|
60 |
+
grayscale_cam = cam(input_tensor=input_img, targets=None)
|
61 |
+
grayscale_cam = grayscale_cam[0, :]
|
62 |
+
|
63 |
+
if gradcam_res:
|
64 |
+
visualization = show_cam_on_image(org_img/255, grayscale_cam, use_rgb=True, image_weight=transparency)
|
65 |
+
else:
|
66 |
+
visualization = org_img
|
67 |
+
|
68 |
+
confidences_sorted = sorted(confidences.items(), key=lambda x: x[1], reverse=True)
|
69 |
+
conf = dict(confidences_sorted[:])
|
70 |
+
confidences_filter = {k: conf[k] for k in list(conf)[:top_res]}
|
71 |
+
|
72 |
+
return classes[prediction[0].item()], visualization, confidences_filter
|
73 |
+
|
74 |
+
title = "CIFAR10 trained on ResNet18 Model with GradCAM"
|
75 |
+
description = "A simple Gradio interface to infer on ResNet model, and get GradCAM results"
|
76 |
+
examples = [["image_01.jpg", True, 3, 0.5, -1], ["image_06.jpg", False, 4, 0.5, -1],
|
77 |
+
["image_02.jpg", True, 5, 0.5, -1], ["image_07.jpg", False,6, 0.65, -1],
|
78 |
+
["image_03.jpg", True, 7, 0.5, -1], ["image_08.jpg", False,8, 0.5, -1],
|
79 |
+
["image_04.jpg", True,3, 0.5, -1], ["image_09.jpg", False,5, 0.5, -1],
|
80 |
+
["image_05.jpg", True,6, 0.77, -1], ["image_10.jpg", False,7, 0.5, -1]]
|
81 |
+
demo = gr.Interface(
|
82 |
+
inference,
|
83 |
+
inputs = [
|
84 |
+
gr.Image(width=256, height=256, label="Input Image"),
|
85 |
+
gr.Checkbox(label="GradCam", info="Check gradcam result?"),
|
86 |
+
gr.Slider(1, 10, value = 3, step=1, label="select number of top predcitions"),
|
87 |
+
gr.Slider(0, 1, value = 0.5, label="Overall Opacity of Image"),
|
88 |
+
gr.Slider(-2, -1, value = -2, step=1, label="Which Layer?")
|
89 |
+
],
|
90 |
+
outputs = [
|
91 |
+
"text",
|
92 |
+
gr.Image(width=256, height=256, label="Output"),
|
93 |
+
gr.Label()
|
94 |
+
],
|
95 |
+
title = title,
|
96 |
+
description = description,
|
97 |
+
examples = examples,
|
98 |
+
)
|
99 |
+
demo.launch()
|