Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
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 custom_resnet import *
|
9 |
+
#from resnet import ResNet18 # Assuming you have a custom ResNet18 implementation
|
10 |
+
|
11 |
+
def load_custom_state_dict(model, state_dict):
|
12 |
+
model_state_dict = model.state_dict()
|
13 |
+
# Filter out unexpected keys
|
14 |
+
filtered_state_dict = {k: v for k, v in state_dict.items() if k in model_state_dict}
|
15 |
+
# Update the model's state_dict
|
16 |
+
model_state_dict.update(filtered_state_dict)
|
17 |
+
# Load the updated state_dict to the model
|
18 |
+
model.load_state_dict(model_state_dict)
|
19 |
+
|
20 |
+
|
21 |
+
model = CustomResNet() # Replace this with your CustomResNet if necessary
|
22 |
+
# Load the state_dict using the custom function
|
23 |
+
state_dict = torch.load("model_pth.ckpt", map_location=torch.device('cpu'))
|
24 |
+
load_custom_state_dict(model, state_dict['state_dict'])
|
25 |
+
|
26 |
+
inv_normalize = transforms.Normalize(
|
27 |
+
mean=[-0.494 / 0.2470, -0.4822 / 0.2435, -0.4465 / 0.2616],
|
28 |
+
std=[1 / 0.2470, 1 / 0.2435, 1 / 0.2616]
|
29 |
+
)
|
30 |
+
classes = ('plane', 'car', 'bird', 'cat', 'deer',
|
31 |
+
'dog', 'frog', 'horse', 'ship', 'truck')
|
32 |
+
|
33 |
+
def inference(input_img, transparency=0.5, target_layer_number=-1, num_images=1, num_top_classes=3):
|
34 |
+
transform = transforms.ToTensor()
|
35 |
+
org_img = input_img
|
36 |
+
input_img = transform(input_img)
|
37 |
+
input_img = input_img.unsqueeze(0)
|
38 |
+
outputs = model(input_img)
|
39 |
+
softmax = torch.nn.Softmax(dim=1)
|
40 |
+
probabilities = softmax(outputs)
|
41 |
+
confidences = {classes[i]: float(probabilities[0, i]) for i in range(10)}
|
42 |
+
|
43 |
+
_, prediction = torch.max(outputs, 1)
|
44 |
+
|
45 |
+
# Get GradCAM for the specified target_layer_number
|
46 |
+
target_layers = [model.layer_2[target_layer_number]]
|
47 |
+
cam = GradCAM(model=model, target_layers=target_layers, use_cuda=False)
|
48 |
+
grayscale_cam = cam(input_tensor=input_img, targets=None)
|
49 |
+
grayscale_cam = grayscale_cam[0, :]
|
50 |
+
img = input_img.squeeze(0)
|
51 |
+
img = inv_normalize(img)
|
52 |
+
rgb_img = np.transpose(img, (1, 2, 0))
|
53 |
+
rgb_img = rgb_img.numpy()
|
54 |
+
|
55 |
+
# Convert org_img (PIL image) to a NumPy array before performing arithmetic operations
|
56 |
+
visualization = show_cam_on_image(org_img / 255, grayscale_cam, use_rgb=True, image_weight=transparency)
|
57 |
+
|
58 |
+
# Create a list to store multiple visualizations
|
59 |
+
|
60 |
+
|
61 |
+
# # Generate multiple GradCAM visualizations if num_images > 1
|
62 |
+
# for _ in range(num_images - 1):
|
63 |
+
# # Get GradCAM for different target_layer_number if provided by the user
|
64 |
+
# if target_layer_number >= -1:
|
65 |
+
# target_layers = [model.layer_2[target_layer_number]]
|
66 |
+
# cam = GradCAM(model=model, target_layers=target_layers, use_cuda=False)
|
67 |
+
# grayscale_cam = cam(input_tensor=input_img, targets=None)
|
68 |
+
# grayscale_cam = grayscale_cam[0, :]
|
69 |
+
|
70 |
+
# visualization = show_cam_on_image(org_img / 255, grayscale_cam, use_rgb=True, image_weight=transparency)
|
71 |
+
# visualizations.append(visualization)
|
72 |
+
|
73 |
+
# Get top classes based on user input (up to a maximum of 10)
|
74 |
+
top_classes = {k: v for k, v in sorted(confidences.items(), key=lambda item: item[1], reverse=True)[:min(num_top_classes, 10)]}
|
75 |
+
|
76 |
+
return top_classes, visualization
|
77 |
+
|
78 |
+
|
79 |
+
|
80 |
+
|
81 |
+
title = "CIFAR10 trained on ResNet18 Model with GradCAM"
|
82 |
+
description = "A simple Gradio interface to infer on ResNet model, and get GradCAM results"
|
83 |
+
examples = [["/content/examples/car_1.jpg",0.5,-1],["/content/examples/car_2.jpg",0.5,-1],["/content/examples/cat_1.jpg",0.5,-1],["/content/examples/cat_2.jpg",0.5,-1],["/content/examples/dog_1.jpg",0.5,-1],["/content/examples/dog_2.jpg",0.5,-1],["/content/examples/frog_1.jpg",0.5,-1],["/content/examples/frog_2.jpg",0.5,-1],["/content/examples/horse_1.jpg",0.5,-1],["/content/examples/horse_2.jpg",0.5,-1]]
|
84 |
+
demo = gr.Interface(
|
85 |
+
inference,
|
86 |
+
inputs = [gr.Image(shape=(32, 32), label="Input Image"),
|
87 |
+
gr.Slider(0, 1, value=0.5, label="Opacity of GradCAM"),
|
88 |
+
gr.Slider(-2, -1, value=-2, step=1, label="Which Layer?"),
|
89 |
+
gr.Number(default=1, label="Number of GradCAM Images to Show"),
|
90 |
+
gr.Slider(1, 10, value=3, step=1, label="Number of Top Classes to Show")],
|
91 |
+
outputs = [gr.Label(num_top_classes=5), gr.Image(shape=(32, 32), label="Output").style(width=128, height=128)],
|
92 |
+
title = title,
|
93 |
+
description = description,
|
94 |
+
examples = examples,
|
95 |
+
)
|
96 |
+
demo.launch()
|