Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,24 +1,6 @@
|
|
1 |
import streamlit as st
|
2 |
|
3 |
-
import
|
4 |
-
from PIL import Image
|
5 |
-
from io import BytesIO
|
6 |
-
from transformers import (
|
7 |
-
AutoModelForImageClassification,
|
8 |
-
AutoFeatureExtractor,
|
9 |
-
AutoConfig,
|
10 |
-
)
|
11 |
-
from torchcam.methods import GradCAM
|
12 |
-
from torchcam.utils import overlay_mask
|
13 |
-
import matplotlib.pyplot as plt
|
14 |
-
from torchvision.transforms.functional import to_pil_image
|
15 |
-
from torchcam import methods
|
16 |
-
|
17 |
-
# TODO I have an error with those
|
18 |
-
# CAM_METHODS = ["CAM", "GradCAM", "GradCAMpp", "SmoothGradCAMpp", "ScoreCAM", "SSCAM", "ISCAM", "XGradCAM", "LayerCAM"]
|
19 |
-
CAM_METHODS = ["CAM", "GradCAM", "GradCAMpp", "LayerCAM"]
|
20 |
-
|
21 |
-
SUPPORTED_MODELS = ["convnext"]
|
22 |
|
23 |
|
24 |
def main():
|
@@ -26,105 +8,7 @@ def main():
|
|
26 |
st.set_page_config(layout="wide")
|
27 |
|
28 |
# Designing the interface
|
29 |
-
st.title("
|
30 |
-
|
31 |
-
# For newline
|
32 |
-
st.write("\n")
|
33 |
-
st.write("`torch-cam`: https://github.com/frgfm/torch-cam")
|
34 |
-
st.write("`transformers`: https://github.com/huggingface/transformers")
|
35 |
-
st.write("Upload an image, select your CAM method and hit the Compute Cam button!")
|
36 |
-
|
37 |
-
# For newline
|
38 |
-
st.write("\n")
|
39 |
-
# Set the columns
|
40 |
-
cols = st.columns((1, 1))
|
41 |
-
cols[0].header("Input image")
|
42 |
-
cols[1].header("Overlayed CAM")
|
43 |
-
# Sidebar
|
44 |
-
# File selection
|
45 |
-
st.sidebar.title("Input selection")
|
46 |
-
# Disabling warning
|
47 |
-
st.set_option("deprecation.showfileUploaderEncoding", False)
|
48 |
-
# Choose your own image
|
49 |
-
uploaded_file = st.sidebar.file_uploader(
|
50 |
-
"Upload files", type=["png", "jpeg", "jpg"]
|
51 |
-
)
|
52 |
-
if uploaded_file is not None:
|
53 |
-
img = Image.open(BytesIO(uploaded_file.read()), mode="r").convert("RGB")
|
54 |
-
else:
|
55 |
-
r = requests.get(
|
56 |
-
"https://i.insider.com/5df126b679d7570ad2044f3e?width=700&format=jpeg&auto=webp"
|
57 |
-
)
|
58 |
-
img = Image.open(BytesIO(r.content))
|
59 |
-
cols[0].image(img, use_column_width=True)
|
60 |
-
|
61 |
-
model_name = st.sidebar.text_input("Model name", "facebook/convnext-tiny-224")
|
62 |
-
|
63 |
-
if model_name is not None:
|
64 |
-
with st.spinner("Loading model..."):
|
65 |
-
config = AutoConfig.from_pretrained(model_name)
|
66 |
-
model_type = config.model_type
|
67 |
-
if model_type not in SUPPORTED_MODELS:
|
68 |
-
st.warning(
|
69 |
-
f"{model_type} not in supported models: {','.join(SUPPORTED_MODELS)}"
|
70 |
-
)
|
71 |
-
else:
|
72 |
-
feature_extractor = AutoFeatureExtractor.from_pretrained(model_name)
|
73 |
-
model = AutoModelForImageClassification.from_pretrained(model_name)
|
74 |
-
|
75 |
-
cam_method = st.sidebar.selectbox("CAM method", CAM_METHODS)
|
76 |
-
if cam_method is not None:
|
77 |
-
cam_extractor = methods.__dict__[cam_method](
|
78 |
-
model, target_layer=model.convnext.encoder.stages[-1].layers[-1]
|
79 |
-
)
|
80 |
-
|
81 |
-
# label choices
|
82 |
-
class_choices = [
|
83 |
-
f"{idx + 1} - {class_name}" for idx, class_name in model.config.id2label.items()
|
84 |
-
]
|
85 |
-
class_selection = st.sidebar.selectbox(
|
86 |
-
"Class selection", ["Predicted class (argmax)"] + class_choices
|
87 |
-
)
|
88 |
-
# for newline
|
89 |
-
st.sidebar.write("\n")
|
90 |
-
|
91 |
-
if st.sidebar.button("Compute CAM"):
|
92 |
-
# compute cam
|
93 |
-
if img is None:
|
94 |
-
st.sidebar.error("Please upload an image first")
|
95 |
-
else:
|
96 |
-
with st.spinner("Analyzing..."):
|
97 |
-
# Set your CAM extractor
|
98 |
-
cam_extractor = GradCAM(
|
99 |
-
model, target_layer=model.convnext.encoder.stages[-1].layers[-1]
|
100 |
-
)
|
101 |
-
inputs = feature_extractor(img, return_tensors="pt")
|
102 |
-
logits = model(**inputs).logits
|
103 |
-
# select the target class
|
104 |
-
if class_selection == "Predicted class (argmax)":
|
105 |
-
class_idx = logits.squeeze(0).argmax().item()
|
106 |
-
else:
|
107 |
-
class_idx = model.config.label2id[
|
108 |
-
class_selection.rpartition(" - ")[-1]
|
109 |
-
]
|
110 |
-
print(class_idx)
|
111 |
-
# run the cam extractor
|
112 |
-
cams = cam_extractor(class_idx, logits)
|
113 |
-
cam = cams[0] if len(cams) == 1 else cam_extractor.fuse_cams(cams)
|
114 |
-
# resize + overlay
|
115 |
-
result = overlay_mask(img, to_pil_image(cam, mode="F"), alpha=0.5)
|
116 |
-
# display it
|
117 |
-
fig, ax = plt.subplots()
|
118 |
-
result = overlay_mask(img, to_pil_image(cam, mode="F"), alpha=0.5)
|
119 |
-
ax.imshow(result)
|
120 |
-
ax.axis("off")
|
121 |
-
cols[1].pyplot(fig)
|
122 |
-
if class_selection == "Predicted class (argmax)":
|
123 |
-
# show the predicted class
|
124 |
-
st.markdown(
|
125 |
-
f"<p style='text align: center'> Predicted class is {config.id2label[class_idx]}</p>",
|
126 |
-
unsafe_allow_html=True,
|
127 |
-
)
|
128 |
-
|
129 |
|
130 |
main()
|
|
|
1 |
import streamlit as st
|
2 |
|
3 |
+
import sys
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
|
6 |
def main():
|
|
|
8 |
st.set_page_config(layout="wide")
|
9 |
|
10 |
# Designing the interface
|
11 |
+
st.title("sys.version)")
|
12 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
main()
|