vilt-b32-finetuned-vqa/app.py

40 lines
1.2 KiB
Python

from transformers import ViltProcessor, ViltForQuestionAnswering
from PIL import Image
import gradio as gr
import torch
from gradio.themes.utils import sizes
theme = gr.themes.Default(radius_size=sizes.radius_none).set(
block_label_text_color = '#4D63FF',
block_title_text_color = '#4D63FF',
button_primary_text_color = '#4D63FF',
button_primary_background_fill='#FFFFFF',
button_primary_border_color='#4D63FF',
button_primary_background_fill_hover='#EDEFFF',
)
processor = ViltProcessor.from_pretrained("dandelin/vilt-b32-finetuned-vqa")
model = ViltForQuestionAnswering.from_pretrained("dandelin/vilt-b32-finetuned-vqa")
def vqa(image, question):
inp = Image.fromarray(image.astype('uint8'), 'RGB')
inputs = processor(inp, question, return_tensors="pt")
outputs = model(**inputs)
logits = outputs.logits
idx = logits.argmax(-1).item()
return model.config.id2label[idx]
demo = gr.Interface(fn=vqa,
inputs=['image', 'text'],
outputs='text',
title = "vqa",
theme=theme,
examples = [['soccer.jpg', 'how many people in the picture?']])
if __name__ == "__main__":
demo.queue(concurrency_count=3).launch()