Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
cbc0399
1
Parent(s):
316d81d
Update app
Browse files- app.py +30 -33
- requirements.txt +0 -2
app.py
CHANGED
@@ -4,7 +4,6 @@ from io import BytesIO
|
|
4 |
from pathlib import Path
|
5 |
|
6 |
import glob
|
7 |
-
import spaces
|
8 |
import numpy as np
|
9 |
import gradio as gr
|
10 |
import rasterio as rio
|
@@ -18,6 +17,11 @@ rcParams["font.size"] = 9
|
|
18 |
rcParams["axes.titlesize"] = 9
|
19 |
IMG_PX = 300
|
20 |
|
|
|
|
|
|
|
|
|
|
|
21 |
EXAMPLES = {
|
22 |
"EuroSAT": {
|
23 |
"images": glob.glob("examples/eurosat/*.tif"),
|
@@ -47,18 +51,6 @@ EXAMPLES = {
|
|
47 |
}
|
48 |
|
49 |
|
50 |
-
def load_eurosat_example():
|
51 |
-
return EXAMPLES["EuroSAT"]["images"], ", ".join(EXAMPLES["EuroSAT"]["classes"])
|
52 |
-
|
53 |
-
|
54 |
-
def load_meterml_example():
|
55 |
-
return EXAMPLES["Meter-ML"]["images"], ", ".join(EXAMPLES["Meter-ML"]["classes"])
|
56 |
-
|
57 |
-
|
58 |
-
def load_terramesh_example():
|
59 |
-
return EXAMPLES["TerraMesh"]["images"], ", ".join(EXAMPLES["TerraMesh"]["classes"])
|
60 |
-
|
61 |
-
|
62 |
pastel1_hex = [mpl.colors.to_hex(c) for c in mpl.colormaps["Pastel1"].colors]
|
63 |
|
64 |
|
@@ -164,8 +156,8 @@ def _bar_chart(top_scores, img_name, cmap) -> str:
|
|
164 |
b64 = base64.b64encode(buf.getvalue()).decode()
|
165 |
return f'<img src="data:image/png;base64,{b64}" style="display:block;margin:auto;width:{IMG_PX}px;" />'
|
166 |
|
167 |
-
|
168 |
-
@spaces.GPU
|
169 |
def classify(images, class_text):
|
170 |
class_names = [c.strip() for c in class_text.split(",") if c.strip()]
|
171 |
cards = []
|
@@ -192,8 +184,25 @@ def classify(images, class_text):
|
|
192 |
)
|
193 |
|
194 |
|
195 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
196 |
|
|
|
197 |
with gr.Blocks(
|
198 |
css="""
|
199 |
.gradio-container
|
@@ -203,7 +212,8 @@ with gr.Blocks(
|
|
203 |
gr.Markdown("## Zero‑shot Classification with Llama3-MS‑CLIP")
|
204 |
gr.Markdown("Provide Sentinel-2 L2A tif files with all 12 bands and define the class names for running zero-shot classification. "
|
205 |
"You can also use S-2 L1C files with 13 bands but the model might not work as well (e.g., misclassifing forests as sea because of the differrently scaled values). "
|
206 |
-
"We provide three sets of example images with class names
|
|
|
207 |
"The images are classified based on the similarity between the images embeddings and text embeddings. "
|
208 |
"You find more information in the [model card](https://huggingface.co/ibm-esa-geospatial/Llama3-MS-CLIP-base) and the [paper](https://arxiv.org/abs/2503.15969). ")
|
209 |
with gr.Row():
|
@@ -212,7 +222,6 @@ with gr.Blocks(
|
|
212 |
)
|
213 |
cls_in = gr.Textbox(
|
214 |
value=", ".join(["Forest", "River", "Buildings", "Agriculture", "Mountain", "Snow"]),
|
215 |
-
# some default classes
|
216 |
label="Class names (comma‑separated)",
|
217 |
)
|
218 |
|
@@ -233,29 +242,17 @@ with gr.Blocks(
|
|
233 |
|
234 |
btn_terramesh.click(
|
235 |
load_terramesh_example,
|
236 |
-
outputs=[img_in, cls_in],
|
237 |
-
).then(
|
238 |
-
classify,
|
239 |
-
inputs=[img_in, cls_in],
|
240 |
-
outputs=out_html,
|
241 |
)
|
242 |
|
243 |
btn_eurosat.click(
|
244 |
load_eurosat_example,
|
245 |
-
outputs=[img_in, cls_in],
|
246 |
-
).then(
|
247 |
-
classify,
|
248 |
-
inputs=[img_in, cls_in],
|
249 |
-
outputs=out_html,
|
250 |
)
|
251 |
|
252 |
btn_meterml.click(
|
253 |
load_meterml_example,
|
254 |
-
outputs=[img_in, cls_in],
|
255 |
-
).then(
|
256 |
-
classify,
|
257 |
-
inputs=[img_in, cls_in],
|
258 |
-
outputs=out_html,
|
259 |
)
|
260 |
|
261 |
if __name__ == "__main__":
|
|
|
4 |
from pathlib import Path
|
5 |
|
6 |
import glob
|
|
|
7 |
import numpy as np
|
8 |
import gradio as gr
|
9 |
import rasterio as rio
|
|
|
17 |
rcParams["axes.titlesize"] = 9
|
18 |
IMG_PX = 300
|
19 |
|
20 |
+
import sys
|
21 |
+
import csv
|
22 |
+
|
23 |
+
csv.field_size_limit(sys.maxsize)
|
24 |
+
|
25 |
EXAMPLES = {
|
26 |
"EuroSAT": {
|
27 |
"images": glob.glob("examples/eurosat/*.tif"),
|
|
|
51 |
}
|
52 |
|
53 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
pastel1_hex = [mpl.colors.to_hex(c) for c in mpl.colormaps["Pastel1"].colors]
|
55 |
|
56 |
|
|
|
156 |
b64 = base64.b64encode(buf.getvalue()).decode()
|
157 |
return f'<img src="data:image/png;base64,{b64}" style="display:block;margin:auto;width:{IMG_PX}px;" />'
|
158 |
|
159 |
+
# import spaces
|
160 |
+
# @spaces.GPU # ZeroGPU does not seem to be working
|
161 |
def classify(images, class_text):
|
162 |
class_names = [c.strip() for c in class_text.split(",") if c.strip()]
|
163 |
cards = []
|
|
|
184 |
)
|
185 |
|
186 |
|
187 |
+
# Cache examples
|
188 |
+
terramesh_html = classify(EXAMPLES["TerraMesh"]["images"], ", ".join(EXAMPLES["TerraMesh"]["classes"]))
|
189 |
+
eurosat_html = classify(EXAMPLES["EuroSAT"]["images"], ", ".join(EXAMPLES["EuroSAT"]["classes"]))
|
190 |
+
meterml_html = classify(EXAMPLES["Meter-ML"]["images"], ", ".join(EXAMPLES["Meter-ML"]["classes"]))
|
191 |
+
|
192 |
+
|
193 |
+
def load_eurosat_example():
|
194 |
+
return EXAMPLES["EuroSAT"]["images"], ", ".join(EXAMPLES["EuroSAT"]["classes"]), eurosat_html
|
195 |
+
|
196 |
+
|
197 |
+
def load_meterml_example():
|
198 |
+
return EXAMPLES["Meter-ML"]["images"], ", ".join(EXAMPLES["Meter-ML"]["classes"]), meterml_html
|
199 |
+
|
200 |
+
|
201 |
+
def load_terramesh_example():
|
202 |
+
return EXAMPLES["TerraMesh"]["images"], ", ".join(EXAMPLES["TerraMesh"]["classes"]), terramesh_html
|
203 |
+
|
204 |
|
205 |
+
# UI
|
206 |
with gr.Blocks(
|
207 |
css="""
|
208 |
.gradio-container
|
|
|
212 |
gr.Markdown("## Zero‑shot Classification with Llama3-MS‑CLIP")
|
213 |
gr.Markdown("Provide Sentinel-2 L2A tif files with all 12 bands and define the class names for running zero-shot classification. "
|
214 |
"You can also use S-2 L1C files with 13 bands but the model might not work as well (e.g., misclassifing forests as sea because of the differrently scaled values). "
|
215 |
+
"We provide three sets of example images with class names and cached outputs. "
|
216 |
+
"The examples are from [EuroSAT](https://arxiv.org/abs/1709.00029), [Meter-ML](https://arxiv.org/abs/2207.11166), and [TerraMesh](https://arxiv.org/abs/2504.11172) (We downloaded S-2 L2A images for the same locations). "
|
217 |
"The images are classified based on the similarity between the images embeddings and text embeddings. "
|
218 |
"You find more information in the [model card](https://huggingface.co/ibm-esa-geospatial/Llama3-MS-CLIP-base) and the [paper](https://arxiv.org/abs/2503.15969). ")
|
219 |
with gr.Row():
|
|
|
222 |
)
|
223 |
cls_in = gr.Textbox(
|
224 |
value=", ".join(["Forest", "River", "Buildings", "Agriculture", "Mountain", "Snow"]),
|
|
|
225 |
label="Class names (comma‑separated)",
|
226 |
)
|
227 |
|
|
|
242 |
|
243 |
btn_terramesh.click(
|
244 |
load_terramesh_example,
|
245 |
+
outputs=[img_in, cls_in, out_html],
|
|
|
|
|
|
|
|
|
246 |
)
|
247 |
|
248 |
btn_eurosat.click(
|
249 |
load_eurosat_example,
|
250 |
+
outputs=[img_in, cls_in, out_html],
|
|
|
|
|
|
|
|
|
251 |
)
|
252 |
|
253 |
btn_meterml.click(
|
254 |
load_meterml_example,
|
255 |
+
outputs=[img_in, cls_in, out_html],
|
|
|
|
|
|
|
|
|
256 |
)
|
257 |
|
258 |
if __name__ == "__main__":
|
requirements.txt
CHANGED
@@ -1,5 +1,3 @@
|
|
1 |
-
--extra-index-url https://download.pytorch.org/whl/cu113
|
2 |
-
torch
|
3 |
gradio>=4.31.0
|
4 |
plotly
|
5 |
rasterio
|
|
|
|
|
|
|
1 |
gradio>=4.31.0
|
2 |
plotly
|
3 |
rasterio
|