visdecode / app.py
martinsinnona
a
db924e3
import gradio as gr
from transformers import AutoProcessor, Pix2StructForConditionalGeneration
import torch
from PIL import Image
import json
import vl_convert as vlc
from io import BytesIO
device = "cuda" if torch.cuda.is_available() else "cpu"
# Load the processor and model
processor = AutoProcessor.from_pretrained("google/matcha-base")
processor.image_processor.is_vqa = False
model = Pix2StructForConditionalGeneration.from_pretrained("martinsinnona/visdecode_B").to(device)
model.eval()
def generate(image):
inputs = processor(images=image, return_tensors="pt", max_patches=1024).to(device)
generated_ids = model.generate(flattened_patches=inputs.flattened_patches, attention_mask=inputs.attention_mask, max_length=600)
generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
# Generate the Vega image
vega = string_to_vega(generated_caption)
vega_image = draw_vega(vega)
return generated_caption, vega_image
def draw_vega(vega, scale=3):
spec = json.dumps(vega, indent=4)
png_data = vlc.vegalite_to_png(vl_spec=spec, scale=scale)
return Image.open(BytesIO(png_data))
def string_to_vega(string):
string = string.replace("'", "\"")
vega = json.loads(string)
for axis in ["x", "y"]:
field = vega["encoding"][axis]["field"]
if field == "":
vega["encoding"][axis]["field"] = axis
vega["encoding"][axis]["title"] = ""
else:
for entry in vega["data"]["values"]:
entry[field] = entry.pop(axis)
return vega
# Create the Gradio interface
iface = gr.Interface(
fn=generate,
inputs=gr.Image(type="pil"),
outputs=[gr.Textbox(),
gr.Image(type="pil", label="Generated Vega Image")],
title="Image to Vega-Lite",
description="Upload an image to generate vega-lite"
)
# Launch the interface
if __name__ == "__main__":
iface.launch(share=True)