Spaces:
Sleeping
Sleeping
martinsinnona
commited on
Commit
•
7b0ea0f
1
Parent(s):
0e2c012
- app.py +39 -9
- requirements.txt +2 -1
app.py
CHANGED
@@ -2,32 +2,62 @@ import gradio as gr
|
|
2 |
from transformers import AutoProcessor, Pix2StructForConditionalGeneration
|
3 |
import torch
|
4 |
from PIL import Image
|
|
|
|
|
|
|
|
|
5 |
|
6 |
# Load the processor and model
|
7 |
processor = AutoProcessor.from_pretrained("google/matcha-base")
|
8 |
processor.image_processor.is_vqa = False
|
9 |
-
|
|
|
10 |
model.eval()
|
11 |
|
12 |
def generate_caption(image):
|
13 |
|
14 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
15 |
-
|
16 |
inputs = processor(images=image, return_tensors="pt", max_patches=1024).to(device)
|
17 |
generated_ids = model.generate(flattened_patches=inputs.flattened_patches, attention_mask=inputs.attention_mask, max_length=600)
|
18 |
generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
# Create the Gradio interface
|
23 |
-
|
|
|
24 |
fn=generate_caption,
|
25 |
inputs=gr.Image(type="pil"),
|
26 |
-
outputs="
|
27 |
-
title="Image to
|
28 |
-
description="Upload an image
|
29 |
)
|
30 |
|
31 |
# Launch the interface
|
32 |
if __name__ == "__main__":
|
33 |
-
|
|
|
2 |
from transformers import AutoProcessor, Pix2StructForConditionalGeneration
|
3 |
import torch
|
4 |
from PIL import Image
|
5 |
+
import json
|
6 |
+
import vl_convert as vlc # Ensure you have this library installed (pip install vl-convert)
|
7 |
+
|
8 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
9 |
|
10 |
# Load the processor and model
|
11 |
processor = AutoProcessor.from_pretrained("google/matcha-base")
|
12 |
processor.image_processor.is_vqa = False
|
13 |
+
|
14 |
+
model = Pix2StructForConditionalGeneration.from_pretrained("martinsinnona/visdecode_B").to(device)
|
15 |
model.eval()
|
16 |
|
17 |
def generate_caption(image):
|
18 |
|
|
|
|
|
19 |
inputs = processor(images=image, return_tensors="pt", max_patches=1024).to(device)
|
20 |
generated_ids = model.generate(flattened_patches=inputs.flattened_patches, attention_mask=inputs.attention_mask, max_length=600)
|
21 |
generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
22 |
+
|
23 |
+
# Generate the Vega image
|
24 |
+
vega = string_to_vega(generated_caption)
|
25 |
+
vega_image = draw_vega(vega)
|
26 |
+
|
27 |
+
return generated_caption, vega_image
|
28 |
+
|
29 |
+
def draw_vega(vega, scale=3):
|
30 |
|
31 |
+
spec = json.dumps(vega, indent=4)
|
32 |
+
png_data = vlc.vegalite_to_png(vl_spec=spec, scale=scale)
|
33 |
+
|
34 |
+
return Image.open(png_data)
|
35 |
+
|
36 |
+
def string_to_vega(string):
|
37 |
+
|
38 |
+
string = string.replace("'", "\"")
|
39 |
+
vega = json.loads(string)
|
40 |
+
|
41 |
+
for axis in ["x", "y"]:
|
42 |
+
field = vega["encoding"][axis]["field"]
|
43 |
+
if field == "":
|
44 |
+
vega["encoding"][axis]["field"] = axis
|
45 |
+
vega["encoding"][axis]["title"] = ""
|
46 |
+
else:
|
47 |
+
for entry in vega["data"]["values"]:
|
48 |
+
entry[field] = entry.pop(axis)
|
49 |
+
return vega
|
50 |
|
51 |
# Create the Gradio interface
|
52 |
+
iface = gr.Interface(
|
53 |
+
|
54 |
fn=generate_caption,
|
55 |
inputs=gr.Image(type="pil"),
|
56 |
+
outputs=[gr.Textbox(), gr.Image(type="pil")],
|
57 |
+
title="Image to Vega-Lite",
|
58 |
+
description="Upload an image to generate vega-lite"
|
59 |
)
|
60 |
|
61 |
# Launch the interface
|
62 |
if __name__ == "__main__":
|
63 |
+
iface.launch(share=True)
|
requirements.txt
CHANGED
@@ -1,2 +1,3 @@
|
|
1 |
transformers
|
2 |
-
torch
|
|
|
|
1 |
transformers
|
2 |
+
torch
|
3 |
+
vl-convert-python
|