fg-mindee
commited on
Commit
•
56fb801
1
Parent(s):
2254e7f
feat: Added Streamlit app
Browse files
app.py
ADDED
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Copyright (C) 2020-2021, François-Guillaume Fernandez.
|
2 |
+
|
3 |
+
# This program is licensed under the Apache License version 2.
|
4 |
+
# See LICENSE or go to <https://www.apache.org/licenses/LICENSE-2.0.txt> for full license details.
|
5 |
+
|
6 |
+
import requests
|
7 |
+
import streamlit as st
|
8 |
+
import matplotlib.pyplot as plt
|
9 |
+
from PIL import Image
|
10 |
+
from io import BytesIO
|
11 |
+
from torchvision import models
|
12 |
+
from torchvision.transforms.functional import resize, to_tensor, normalize, to_pil_image
|
13 |
+
|
14 |
+
from torchcam import cams
|
15 |
+
from torchcam.utils import overlay_mask
|
16 |
+
|
17 |
+
|
18 |
+
CAM_METHODS = ["CAM", "GradCAM", "GradCAMpp", "SmoothGradCAMpp", "ScoreCAM", "SSCAM", "ISCAM", "XGradCAM", "LayerCAM"]
|
19 |
+
TV_MODELS = ["resnet18", "resnet50", "mobilenet_v2", "mobilenet_v3_small", "mobilenet_v3_large"]
|
20 |
+
LABEL_MAP = requests.get(
|
21 |
+
"https://raw.githubusercontent.com/anishathalye/imagenet-simple-labels/master/imagenet-simple-labels.json"
|
22 |
+
).json()
|
23 |
+
|
24 |
+
|
25 |
+
def main():
|
26 |
+
|
27 |
+
# Wide mode
|
28 |
+
st.set_page_config(layout="wide")
|
29 |
+
|
30 |
+
# Designing the interface
|
31 |
+
st.title("TorchCAM: class activation explorer")
|
32 |
+
# For newline
|
33 |
+
st.write('\n')
|
34 |
+
# Set the columns
|
35 |
+
cols = st.columns((1, 1, 1))
|
36 |
+
cols[0].header("Input image")
|
37 |
+
cols[1].header("Raw CAM")
|
38 |
+
cols[-1].header("Overlayed CAM")
|
39 |
+
|
40 |
+
# Sidebar
|
41 |
+
# File selection
|
42 |
+
st.sidebar.title("Input selection")
|
43 |
+
# Disabling warning
|
44 |
+
st.set_option('deprecation.showfileUploaderEncoding', False)
|
45 |
+
# Choose your own image
|
46 |
+
uploaded_file = st.sidebar.file_uploader("Upload files", type=['png', 'jpeg', 'jpg'])
|
47 |
+
if uploaded_file is not None:
|
48 |
+
img = Image.open(BytesIO(uploaded_file.read()), mode='r').convert('RGB')
|
49 |
+
|
50 |
+
cols[0].image(img, use_column_width=True)
|
51 |
+
|
52 |
+
# Model selection
|
53 |
+
st.sidebar.title("Setup")
|
54 |
+
tv_model = st.sidebar.selectbox("Classification model", TV_MODELS)
|
55 |
+
default_layer = ""
|
56 |
+
if tv_model is not None:
|
57 |
+
with st.spinner('Loading model...'):
|
58 |
+
model = models.__dict__[tv_model](pretrained=True).eval()
|
59 |
+
default_layer = cams.utils.locate_candidate_layer(model, (3, 224, 224))
|
60 |
+
|
61 |
+
target_layer = st.sidebar.text_input("Target layer", default_layer)
|
62 |
+
cam_method = st.sidebar.selectbox("CAM method", CAM_METHODS)
|
63 |
+
if cam_method is not None:
|
64 |
+
cam_extractor = cams.__dict__[cam_method](
|
65 |
+
model,
|
66 |
+
target_layer=target_layer if len(target_layer) > 0 else None
|
67 |
+
)
|
68 |
+
|
69 |
+
class_choices = [f"{idx + 1} - {class_name}" for idx, class_name in enumerate(LABEL_MAP)]
|
70 |
+
class_selection = st.sidebar.selectbox("Class selection", ["Predicted class (argmax)"] + class_choices)
|
71 |
+
|
72 |
+
# For newline
|
73 |
+
st.sidebar.write('\n')
|
74 |
+
|
75 |
+
if st.sidebar.button("Compute CAM"):
|
76 |
+
|
77 |
+
if uploaded_file is None:
|
78 |
+
st.sidebar.error("Please upload an image first")
|
79 |
+
|
80 |
+
else:
|
81 |
+
with st.spinner('Analyzing...'):
|
82 |
+
|
83 |
+
# Preprocess image
|
84 |
+
img_tensor = normalize(to_tensor(resize(img, (224, 224))), [0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
|
85 |
+
|
86 |
+
# Forward the image to the model
|
87 |
+
out = model(img_tensor.unsqueeze(0))
|
88 |
+
# Select the target class
|
89 |
+
if class_selection == "Predicted class (argmax)":
|
90 |
+
class_idx = out.squeeze(0).argmax().item()
|
91 |
+
else:
|
92 |
+
class_idx = LABEL_MAP.index(class_selection.rpartition(" - ")[-1])
|
93 |
+
# Retrieve the CAM
|
94 |
+
activation_map = cam_extractor(class_idx, out)[0]
|
95 |
+
# Plot the raw heatmap
|
96 |
+
fig, ax = plt.subplots()
|
97 |
+
ax.imshow(activation_map.numpy())
|
98 |
+
ax.axis('off')
|
99 |
+
cols[1].pyplot(fig)
|
100 |
+
|
101 |
+
# Overlayed CAM
|
102 |
+
fig, ax = plt.subplots()
|
103 |
+
result = overlay_mask(img, to_pil_image(activation_map, mode='F'), alpha=0.5)
|
104 |
+
ax.imshow(result)
|
105 |
+
ax.axis('off')
|
106 |
+
cols[-1].pyplot(fig)
|
107 |
+
|
108 |
+
|
109 |
+
if __name__ == '__main__':
|
110 |
+
main()
|