F-G Fernandez
commited on
Commit
•
c59b75f
1
Parent(s):
a9a3664
fix: Fixed matplotlib call
Browse files
app.py
CHANGED
@@ -1,21 +1,22 @@
|
|
1 |
-
# Copyright (C)
|
2 |
|
3 |
-
# This program is licensed under the Apache License
|
4 |
-
# See LICENSE or go to <https://www.apache.org/licenses/LICENSE-2.0
|
5 |
|
|
|
|
|
|
|
6 |
import requests
|
7 |
import streamlit as st
|
8 |
-
import
|
9 |
from PIL import Image
|
10 |
-
from io import BytesIO
|
11 |
from torchvision import models
|
12 |
-
from torchvision.transforms.functional import
|
13 |
|
14 |
from torchcam import methods
|
15 |
from torchcam.methods._utils import locate_candidate_layer
|
16 |
from torchcam.utils import overlay_mask
|
17 |
|
18 |
-
|
19 |
CAM_METHODS = ["CAM", "GradCAM", "GradCAMpp", "SmoothGradCAMpp", "ScoreCAM", "SSCAM", "ISCAM", "XGradCAM", "LayerCAM"]
|
20 |
TV_MODELS = [
|
21 |
"resnet18",
|
@@ -39,10 +40,7 @@ def main():
|
|
39 |
# Designing the interface
|
40 |
st.title("TorchCAM: class activation explorer")
|
41 |
# For newline
|
42 |
-
st.write(
|
43 |
-
st.write('Check the project at: https://github.com/frgfm/torch-cam')
|
44 |
-
# For newline
|
45 |
-
st.write('\n')
|
46 |
# Set the columns
|
47 |
cols = st.columns((1, 1, 1))
|
48 |
cols[0].header("Input image")
|
@@ -53,36 +51,50 @@ def main():
|
|
53 |
# File selection
|
54 |
st.sidebar.title("Input selection")
|
55 |
# Disabling warning
|
56 |
-
st.set_option(
|
57 |
# Choose your own image
|
58 |
-
uploaded_file = st.sidebar.file_uploader("Upload files", type=[
|
59 |
if uploaded_file is not None:
|
60 |
-
img = Image.open(BytesIO(uploaded_file.read()), mode=
|
61 |
|
62 |
cols[0].image(img, use_column_width=True)
|
63 |
|
64 |
# Model selection
|
65 |
st.sidebar.title("Setup")
|
66 |
-
tv_model = st.sidebar.selectbox(
|
|
|
|
|
|
|
|
|
67 |
default_layer = ""
|
68 |
if tv_model is not None:
|
69 |
-
with st.spinner(
|
70 |
model = models.__dict__[tv_model](pretrained=True).eval()
|
71 |
default_layer = locate_candidate_layer(model, (3, 224, 224))
|
72 |
|
73 |
-
|
74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
if cam_method is not None:
|
76 |
cam_extractor = methods.__dict__[cam_method](
|
77 |
-
model,
|
78 |
-
target_layer=target_layer.split("+") if len(target_layer) > 0 else None
|
79 |
)
|
80 |
|
81 |
class_choices = [f"{idx + 1} - {class_name}" for idx, class_name in enumerate(LABEL_MAP)]
|
82 |
class_selection = st.sidebar.selectbox("Class selection", ["Predicted class (argmax)"] + class_choices)
|
83 |
|
84 |
# For newline
|
85 |
-
st.sidebar.write(
|
86 |
|
87 |
if st.sidebar.button("Compute CAM"):
|
88 |
|
@@ -90,11 +102,14 @@ def main():
|
|
90 |
st.sidebar.error("Please upload an image first")
|
91 |
|
92 |
else:
|
93 |
-
with st.spinner(
|
94 |
|
95 |
# Preprocess image
|
96 |
img_tensor = normalize(to_tensor(resize(img, (224, 224))), [0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
|
97 |
|
|
|
|
|
|
|
98 |
# Forward the image to the model
|
99 |
out = model(img_tensor.unsqueeze(0))
|
100 |
# Select the target class
|
@@ -103,22 +118,22 @@ def main():
|
|
103 |
else:
|
104 |
class_idx = LABEL_MAP.index(class_selection.rpartition(" - ")[-1])
|
105 |
# Retrieve the CAM
|
106 |
-
|
107 |
# Fuse the CAMs if there are several
|
108 |
-
|
109 |
# Plot the raw heatmap
|
110 |
fig, ax = plt.subplots()
|
111 |
-
ax.imshow(
|
112 |
-
ax.axis(
|
113 |
cols[1].pyplot(fig)
|
114 |
|
115 |
# Overlayed CAM
|
116 |
fig, ax = plt.subplots()
|
117 |
-
result = overlay_mask(img, to_pil_image(
|
118 |
ax.imshow(result)
|
119 |
-
ax.axis(
|
120 |
cols[-1].pyplot(fig)
|
121 |
|
122 |
|
123 |
-
if __name__ ==
|
124 |
main()
|
|
|
1 |
+
# Copyright (C) 2021-2022, François-Guillaume Fernandez.
|
2 |
|
3 |
+
# This program is licensed under the Apache License 2.0.
|
4 |
+
# See LICENSE or go to <https://www.apache.org/licenses/LICENSE-2.0> for full license details.
|
5 |
|
6 |
+
from io import BytesIO
|
7 |
+
|
8 |
+
import matplotlib.pyplot as plt
|
9 |
import requests
|
10 |
import streamlit as st
|
11 |
+
import torch
|
12 |
from PIL import Image
|
|
|
13 |
from torchvision import models
|
14 |
+
from torchvision.transforms.functional import normalize, resize, to_pil_image, to_tensor
|
15 |
|
16 |
from torchcam import methods
|
17 |
from torchcam.methods._utils import locate_candidate_layer
|
18 |
from torchcam.utils import overlay_mask
|
19 |
|
|
|
20 |
CAM_METHODS = ["CAM", "GradCAM", "GradCAMpp", "SmoothGradCAMpp", "ScoreCAM", "SSCAM", "ISCAM", "XGradCAM", "LayerCAM"]
|
21 |
TV_MODELS = [
|
22 |
"resnet18",
|
|
|
40 |
# Designing the interface
|
41 |
st.title("TorchCAM: class activation explorer")
|
42 |
# For newline
|
43 |
+
st.write("\n")
|
|
|
|
|
|
|
44 |
# Set the columns
|
45 |
cols = st.columns((1, 1, 1))
|
46 |
cols[0].header("Input image")
|
|
|
51 |
# File selection
|
52 |
st.sidebar.title("Input selection")
|
53 |
# Disabling warning
|
54 |
+
st.set_option("deprecation.showfileUploaderEncoding", False)
|
55 |
# Choose your own image
|
56 |
+
uploaded_file = st.sidebar.file_uploader("Upload files", type=["png", "jpeg", "jpg"])
|
57 |
if uploaded_file is not None:
|
58 |
+
img = Image.open(BytesIO(uploaded_file.read()), mode="r").convert("RGB")
|
59 |
|
60 |
cols[0].image(img, use_column_width=True)
|
61 |
|
62 |
# Model selection
|
63 |
st.sidebar.title("Setup")
|
64 |
+
tv_model = st.sidebar.selectbox(
|
65 |
+
"Classification model",
|
66 |
+
TV_MODELS,
|
67 |
+
help="Supported models from Torchvision",
|
68 |
+
)
|
69 |
default_layer = ""
|
70 |
if tv_model is not None:
|
71 |
+
with st.spinner("Loading model..."):
|
72 |
model = models.__dict__[tv_model](pretrained=True).eval()
|
73 |
default_layer = locate_candidate_layer(model, (3, 224, 224))
|
74 |
|
75 |
+
if torch.cuda.is_available():
|
76 |
+
model = model.cuda()
|
77 |
+
|
78 |
+
target_layer = st.sidebar.text_input(
|
79 |
+
"Target layer",
|
80 |
+
default_layer,
|
81 |
+
help='If you want to target several layers, add a "+" separator (e.g. "layer3+layer4")',
|
82 |
+
)
|
83 |
+
cam_method = st.sidebar.selectbox(
|
84 |
+
"CAM method",
|
85 |
+
CAM_METHODS,
|
86 |
+
help="The way your class activation map will be computed",
|
87 |
+
)
|
88 |
if cam_method is not None:
|
89 |
cam_extractor = methods.__dict__[cam_method](
|
90 |
+
model, target_layer=[s.strip() for s in target_layer.split("+")] if len(target_layer) > 0 else None
|
|
|
91 |
)
|
92 |
|
93 |
class_choices = [f"{idx + 1} - {class_name}" for idx, class_name in enumerate(LABEL_MAP)]
|
94 |
class_selection = st.sidebar.selectbox("Class selection", ["Predicted class (argmax)"] + class_choices)
|
95 |
|
96 |
# For newline
|
97 |
+
st.sidebar.write("\n")
|
98 |
|
99 |
if st.sidebar.button("Compute CAM"):
|
100 |
|
|
|
102 |
st.sidebar.error("Please upload an image first")
|
103 |
|
104 |
else:
|
105 |
+
with st.spinner("Analyzing..."):
|
106 |
|
107 |
# Preprocess image
|
108 |
img_tensor = normalize(to_tensor(resize(img, (224, 224))), [0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
|
109 |
|
110 |
+
if torch.cuda.is_available():
|
111 |
+
img_tensor = img_tensor.cuda()
|
112 |
+
|
113 |
# Forward the image to the model
|
114 |
out = model(img_tensor.unsqueeze(0))
|
115 |
# Select the target class
|
|
|
118 |
else:
|
119 |
class_idx = LABEL_MAP.index(class_selection.rpartition(" - ")[-1])
|
120 |
# Retrieve the CAM
|
121 |
+
act_maps = cam_extractor(class_idx, out)
|
122 |
# Fuse the CAMs if there are several
|
123 |
+
activation_map = act_maps[0] if len(act_maps) == 1 else cam_extractor.fuse_cams(act_maps)
|
124 |
# Plot the raw heatmap
|
125 |
fig, ax = plt.subplots()
|
126 |
+
ax.imshow(activation_map.squeeze(0).cpu().numpy())
|
127 |
+
ax.axis("off")
|
128 |
cols[1].pyplot(fig)
|
129 |
|
130 |
# Overlayed CAM
|
131 |
fig, ax = plt.subplots()
|
132 |
+
result = overlay_mask(img, to_pil_image(activation_map, mode="F"), alpha=0.5)
|
133 |
ax.imshow(result)
|
134 |
+
ax.axis("off")
|
135 |
cols[-1].pyplot(fig)
|
136 |
|
137 |
|
138 |
+
if __name__ == "__main__":
|
139 |
main()
|