app.py

import gradio as gr
from transformers import AutoProcessor, AutoTokenizer, AutoImageProcessor, AutoModelForCausalLM, BlipForConditionalGeneration, VisionEncoderDecoderModel
import torch

git_processor_base = AutoProcessor.from_pretrained("microsoft/git-base-coco")
git_model_base = AutoModelForCausalLM.from_pretrained("microsoft/git-base-coco")

git_processor_large = AutoProcessor.from_pretrained("microsoft/git-large-coco")
git_model_large = AutoModelForCausalLM.from_pretrained("microsoft/git-large-coco")

blip_processor_base = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
blip_model_base = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")

blip_processor_large = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
blip_model_large = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")

vitgpt_processor = AutoImageProcessor.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
vitgpt_model = VisionEncoderDecoderModel.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
vitgpt_tokenizer = AutoTokenizer.from_pretrained("nlpconnect/vit-gpt2-image-captioning")

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

git_model_base.to(device)
blip_model_base.to(device)
git_model_large.to(device)
blip_model_large.to(device)
vitgpt_model.to(device)

def generate_caption(processor, model, image, tokenizer=None):
    inputs = processor(images=image, return_tensors="pt").to(device)
    
    generated_ids = model.generate(pixel_values=inputs.pixel_values, max_length=50)

    if tokenizer is not None:
        generated_caption = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
    else:
        generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
   
    return generated_caption


def generate_captions(image):
    caption_git_base = generate_caption(git_processor_base, git_model_base, image)

    caption_git_large = generate_caption(git_processor_large, git_model_large, image)

    caption_blip_base = generate_caption(blip_processor_base, blip_model_base, image)

    caption_blip_large = generate_caption(blip_processor_large, blip_model_large, image)

    caption_vitgpt = generate_caption(vitgpt_processor, vitgpt_model, image, vitgpt_tokenizer)

    return caption_git_base, caption_git_large, caption_blip_base, caption_blip_large, caption_vitgpt

   
examples = [["test-1.jpeg"], ["test-2.jpeg"], ["test-3.jpeg"], ["test-4.jpeg"], ["test-5.jpeg"], ["test-6.jpg"]]
outputs = [gr.outputs.Textbox(label="Caption generated by GIT-base"), gr.outputs.Textbox(label="Caption generated by GIT-large"), gr.outputs.Textbox(label="Caption generated by BLIP-base"), gr.outputs.Textbox(label="Caption generated by BLIP-large"), gr.outputs.Textbox(label="Caption generated by ViT+GPT-2")] 


interface = gr.Interface(fn=generate_captions, 
                         inputs=gr.inputs.Image(type="pil"),
                         outputs=outputs,
                         examples=examples,  
                         enable_queue=True)
interface.launch(debug=True)