app.py

import gradio as gr
import sys
from BLIP.models.blip import blip_decoder
from PIL import Image
import requests
import torch
from torchvision import transforms
from torchvision.transforms.functional import InterpolationMode
from urllib.parse import urlparse

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

image_size = 384
transform = transforms.Compose([
        transforms.ToTensor(),
        transforms.Resize((image_size,image_size),interpolation=InterpolationMode.BICUBIC),
        transforms.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
        ]) 

model_url = "https://technionmail-my.sharepoint.com/personal/snoamr_campus_technion_ac_il/_layouts/15/download.aspx?share=EZxgXQaBXGREgDsQiaTcwAAB0z8jQA_hgAnwwPQDt8Dgew"
model = blip_decoder(pretrained=model_url, image_size=384, vit='base')
model.eval()
model = model.to(device)

def inference(raw_image):
    # raw_image = torch.tensor(raw_image)
    image = transform(raw_image).unsqueeze(0).to(device)
    with torch.no_grad():
          caption = model.generate(image, sample=False, num_beams=3, max_length=60, min_length=5) 
    return caption[0]


inputs = [gr.Image(type='pil', interactive=False),]
outputs = gr.outputs.Textbox(label="Caption")

description = "Gradio demo for FuseCap: Leveraging Large Language Models to Fuse Visual Data into Enriched Image Captions. This demo features a BLIP-based model, trained using FuseCap."
examples = [["surfer.jpg"], ["bike.jpg"]]
article = "<p style='text-align: center'><a href='google.com' target='_blank'>place holder</a>"


iface = gr.Interface(fn=inference, 
                    inputs="image",
                    outputs="text", 
                    title="FuseCap",
                    description=description,
                    article=article,
                    examples=examples,
                    enable_queue=True)
iface.launch()