Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
d00bf12
1
Parent(s):
626932d
add app.py and requirements.txt
Browse files- app.py +263 -0
- requirements.txt +10 -0
app.py
ADDED
@@ -0,0 +1,263 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import time
|
2 |
+
|
3 |
+
import gradio as gr
|
4 |
+
import matplotlib.cm as cm
|
5 |
+
import numpy as np
|
6 |
+
import plotly.graph_objects as go
|
7 |
+
import spaces
|
8 |
+
import torch
|
9 |
+
from PIL import Image
|
10 |
+
from transformers import AutoImageProcessor, AutoModelForKeypointMatching
|
11 |
+
from transformers.image_utils import to_numpy_array
|
12 |
+
|
13 |
+
|
14 |
+
@spaces.GPU
|
15 |
+
def process_images(image1, image2, model_name):
|
16 |
+
"""
|
17 |
+
Process two images and return a plot of the matching keypoints.
|
18 |
+
"""
|
19 |
+
if image1 is None or image2 is None:
|
20 |
+
return None
|
21 |
+
|
22 |
+
images = [image1, image2]
|
23 |
+
|
24 |
+
processor = AutoImageProcessor.from_pretrained(model_name)
|
25 |
+
model = AutoModelForKeypointMatching.from_pretrained(model_name, device_map="auto")
|
26 |
+
inputs = processor(images, return_tensors="pt")
|
27 |
+
inputs = inputs.to(model.device)
|
28 |
+
print(
|
29 |
+
f"Model {model_name} is on device: {model.device} and inputs are on device: {inputs['pixel_values'].device}"
|
30 |
+
)
|
31 |
+
|
32 |
+
with torch.no_grad():
|
33 |
+
outputs = model(**inputs)
|
34 |
+
|
35 |
+
image_sizes = [[(image.height, image.width) for image in images]]
|
36 |
+
outputs = processor.post_process_keypoint_matching(
|
37 |
+
outputs, image_sizes, threshold=0.2
|
38 |
+
)
|
39 |
+
output = outputs[0]
|
40 |
+
|
41 |
+
image1 = to_numpy_array(image1)
|
42 |
+
image2 = to_numpy_array(image2)
|
43 |
+
|
44 |
+
height0, width0 = image1.shape[:2]
|
45 |
+
height1, width1 = image2.shape[:2]
|
46 |
+
|
47 |
+
# Create PIL image from numpy array
|
48 |
+
pil_img = Image.fromarray((image1 / 255.0 * 255).astype(np.uint8))
|
49 |
+
pil_img2 = Image.fromarray((image2 / 255.0 * 255).astype(np.uint8))
|
50 |
+
|
51 |
+
fig = go.Figure()
|
52 |
+
|
53 |
+
# Create colormap (red-yellow-green: red for low scores, green for high scores)
|
54 |
+
colormap = cm.RdYlGn
|
55 |
+
|
56 |
+
# Get keypoints
|
57 |
+
keypoints0_x, keypoints0_y = output["keypoints0"].unbind(1)
|
58 |
+
keypoints1_x, keypoints1_y = output["keypoints1"].unbind(1)
|
59 |
+
|
60 |
+
# Add a separate trace for each match (line + markers) to enable highlighting
|
61 |
+
for keypoint0_x, keypoint0_y, keypoint1_x, keypoint1_y, matching_score in zip(
|
62 |
+
keypoints0_x,
|
63 |
+
keypoints0_y,
|
64 |
+
keypoints1_x,
|
65 |
+
keypoints1_y,
|
66 |
+
output["matching_scores"],
|
67 |
+
):
|
68 |
+
color_val = matching_score.item()
|
69 |
+
rgba_color = colormap(color_val)
|
70 |
+
|
71 |
+
# Convert to rgba string with transparency
|
72 |
+
color = f"rgba({int(rgba_color[0] * 255)}, {int(rgba_color[1] * 255)}, {int(rgba_color[2] * 255)}, 0.8)"
|
73 |
+
|
74 |
+
hover_text = (
|
75 |
+
f"Score: {matching_score.item():.3f}<br>"
|
76 |
+
f"Point 1: ({keypoint0_x.item():.1f}, {keypoint0_y.item():.1f})<br>"
|
77 |
+
f"Point 2: ({keypoint1_x.item():.1f}, {keypoint1_y.item():.1f})"
|
78 |
+
)
|
79 |
+
|
80 |
+
fig.add_trace(
|
81 |
+
go.Scatter(
|
82 |
+
x=[keypoint0_x.item(), keypoint1_x.item() + width0],
|
83 |
+
y=[keypoint0_y.item(), keypoint1_y.item()],
|
84 |
+
mode="lines+markers",
|
85 |
+
line=dict(color=color, width=2),
|
86 |
+
marker=dict(color=color, size=5, opacity=0.8),
|
87 |
+
hoverinfo="text",
|
88 |
+
hovertext=hover_text,
|
89 |
+
showlegend=False,
|
90 |
+
)
|
91 |
+
)
|
92 |
+
|
93 |
+
# Update layout to use images as background
|
94 |
+
fig.update_layout(
|
95 |
+
xaxis=dict(
|
96 |
+
range=[0, width0 + width1],
|
97 |
+
showgrid=False,
|
98 |
+
zeroline=False,
|
99 |
+
showticklabels=False,
|
100 |
+
),
|
101 |
+
yaxis=dict(
|
102 |
+
range=[max(height0, height1), 0],
|
103 |
+
showgrid=False,
|
104 |
+
zeroline=False,
|
105 |
+
showticklabels=False,
|
106 |
+
scaleanchor="x",
|
107 |
+
scaleratio=1,
|
108 |
+
),
|
109 |
+
margin=dict(l=0, r=0, t=0, b=0),
|
110 |
+
autosize=True,
|
111 |
+
images=[
|
112 |
+
dict(
|
113 |
+
source=pil_img,
|
114 |
+
xref="x",
|
115 |
+
yref="y",
|
116 |
+
x=0,
|
117 |
+
y=0,
|
118 |
+
sizex=width0,
|
119 |
+
sizey=height0,
|
120 |
+
sizing="stretch",
|
121 |
+
opacity=1,
|
122 |
+
layer="below",
|
123 |
+
),
|
124 |
+
dict(
|
125 |
+
source=pil_img2,
|
126 |
+
xref="x",
|
127 |
+
yref="y",
|
128 |
+
x=width0,
|
129 |
+
y=0,
|
130 |
+
sizex=width1,
|
131 |
+
sizey=height1,
|
132 |
+
sizing="stretch",
|
133 |
+
opacity=1,
|
134 |
+
layer="below",
|
135 |
+
),
|
136 |
+
],
|
137 |
+
)
|
138 |
+
|
139 |
+
return fig
|
140 |
+
|
141 |
+
|
142 |
+
# Create the Gradio interface
|
143 |
+
with gr.Blocks(title="EfficientLoFTR Matching Demo") as demo:
|
144 |
+
gr.Markdown("# EfficientLoFTR Matching Demo")
|
145 |
+
gr.Markdown(
|
146 |
+
"Upload two images and get a side-by-side matching of your images using EfficientLoFTR."
|
147 |
+
)
|
148 |
+
gr.Markdown("""
|
149 |
+
## How to use:
|
150 |
+
1. Select an EfficientLoFTR model (Original EfficientLoFTR or MatchAnything)
|
151 |
+
2. Upload two images using the file uploaders below
|
152 |
+
3. Click the 'Match Images' button
|
153 |
+
4. View the matched output image below. Higher scores are green, lower scores are red.
|
154 |
+
|
155 |
+
The app will create a side-by-side matching of your images using EfficientLoFTR.
|
156 |
+
You can also select an example image pair from the dataset below.
|
157 |
+
""")
|
158 |
+
|
159 |
+
with gr.Row():
|
160 |
+
# Detector choice selector
|
161 |
+
detector_choice = gr.Radio(
|
162 |
+
choices=[("Original EfficientLoFTR", "zju-community/efficientloftr"), ("MatchAnything", "zju-community/matchanything_eloftr")],
|
163 |
+
value="Original EfficientLoFTR",
|
164 |
+
label="EfficientLoFTR Model",
|
165 |
+
info="Choose between original EfficientLoFTR or MatchAnything"
|
166 |
+
)
|
167 |
+
|
168 |
+
with gr.Row():
|
169 |
+
# Input images on the same row
|
170 |
+
image1 = gr.Image(label="First Image", type="pil")
|
171 |
+
image2 = gr.Image(label="Second Image", type="pil")
|
172 |
+
|
173 |
+
# Process button
|
174 |
+
process_btn = gr.Button("Match Images", variant="primary")
|
175 |
+
|
176 |
+
# Output plot
|
177 |
+
output_plot = gr.Plot(label="Matching Results", scale=2)
|
178 |
+
|
179 |
+
# Connect the function
|
180 |
+
process_btn.click(fn=process_images, inputs=[image1, image2, detector_choice], outputs=[output_plot])
|
181 |
+
|
182 |
+
# Add some example usage
|
183 |
+
|
184 |
+
examples = gr.Dataset(
|
185 |
+
components=[image1, image2],
|
186 |
+
label="Example Image Pairs",
|
187 |
+
samples=[
|
188 |
+
[
|
189 |
+
"https://raw.githubusercontent.com/magicleap/SuperGluePretrainedNetwork/refs/heads/master/assets/phototourism_sample_images/united_states_capitol_98169888_3347710852.jpg",
|
190 |
+
"https://raw.githubusercontent.com/magicleap/SuperGluePretrainedNetwork/refs/heads/master/assets/phototourism_sample_images/united_states_capitol_26757027_6717084061.jpg",
|
191 |
+
],
|
192 |
+
[
|
193 |
+
"https://raw.githubusercontent.com/cvg/LightGlue/refs/heads/main/assets/DSC_0410.JPG",
|
194 |
+
"https://raw.githubusercontent.com/cvg/LightGlue/refs/heads/main/assets/DSC_0411.JPG",
|
195 |
+
],
|
196 |
+
[
|
197 |
+
"https://raw.githubusercontent.com/cvg/LightGlue/refs/heads/main/assets/sacre_coeur1.jpg",
|
198 |
+
"https://raw.githubusercontent.com/cvg/LightGlue/refs/heads/main/assets/sacre_coeur2.jpg",
|
199 |
+
],
|
200 |
+
[
|
201 |
+
"https://raw.githubusercontent.com/magicleap/SuperGluePretrainedNetwork/refs/heads/master/assets/phototourism_sample_images/piazza_san_marco_06795901_3725050516.jpg",
|
202 |
+
"https://raw.githubusercontent.com/magicleap/SuperGluePretrainedNetwork/refs/heads/master/assets/phototourism_sample_images/piazza_san_marco_58751010_4849458397.jpg",
|
203 |
+
],
|
204 |
+
[
|
205 |
+
"https://raw.githubusercontent.com/magicleap/SuperGluePretrainedNetwork/refs/heads/master/assets/phototourism_sample_images/london_bridge_19481797_2295892421.jpg",
|
206 |
+
"https://raw.githubusercontent.com/magicleap/SuperGluePretrainedNetwork/refs/heads/master/assets/phototourism_sample_images/london_bridge_78916675_4568141288.jpg",
|
207 |
+
],
|
208 |
+
# MatchAnything multi-modality pairs
|
209 |
+
[
|
210 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/MTV_thermal_vis_1.jpg",
|
211 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/MTV_thermal_vis_2.jpg",
|
212 |
+
],
|
213 |
+
[
|
214 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/MTV_thermal_vis_pair2_1.jpg",
|
215 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/MTV_thermal_vis_pair2_2.jpg",
|
216 |
+
],
|
217 |
+
[
|
218 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/ct_mr_1.png",
|
219 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/ct_mr_2.png",
|
220 |
+
],
|
221 |
+
[
|
222 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/mri_ut_1.jpg",
|
223 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/mri_ut_2.jpg",
|
224 |
+
],
|
225 |
+
[
|
226 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/rgb_2.png",
|
227 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/robot_real_world_2.png",
|
228 |
+
],
|
229 |
+
[
|
230 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/robot_render_1.png",
|
231 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/robot_real_world_2.png",
|
232 |
+
],
|
233 |
+
[
|
234 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/thermal_1.jpg",
|
235 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/thermal_vis_1.jpg",
|
236 |
+
],
|
237 |
+
[
|
238 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/thermal_vis_1.jpg",
|
239 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/thermal_vis_2.jpg",
|
240 |
+
],
|
241 |
+
[
|
242 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/vis_event_1.png",
|
243 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/vis_event_2.png",
|
244 |
+
],
|
245 |
+
[
|
246 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/vis_map_1.jpg",
|
247 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/vis_map_2.jpg",
|
248 |
+
],
|
249 |
+
[
|
250 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/vis_map_pair2_1.jpg",
|
251 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/vis_map_pair2_2.jpg",
|
252 |
+
],
|
253 |
+
[
|
254 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/vis_thermal_ground_1.png",
|
255 |
+
"https://huggingface.co/spaces/LittleFrog/MatchAnything/resolve/main/imcui/datasets/multi_modality_pairs/vis_thermal_ground_2.png",
|
256 |
+
],
|
257 |
+
],
|
258 |
+
)
|
259 |
+
|
260 |
+
examples.select(lambda x: (x[0], x[1]), [examples], [image1, image2])
|
261 |
+
|
262 |
+
if __name__ == "__main__":
|
263 |
+
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
gradio>=5.34.2
|
2 |
+
Pillow>=10.0.0
|
3 |
+
numpy>=1.24.0
|
4 |
+
transformers @ git+https://github.com/huggingface/transformers.git@52aaa3f5004d18ecb148c82534eb9eec8ac20f8f
|
5 |
+
matplotlib
|
6 |
+
torch
|
7 |
+
plotly
|
8 |
+
spaces
|
9 |
+
accelerate
|
10 |
+
kornia
|