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

torch.hub.download_url_to_file('http://images.cocodataset.org/val2017/000000039769.jpg', 'cats.jpg')
torch.hub.download_url_to_file('https://huggingface.co/datasets/nielsr/textcaps-sample/resolve/main/stop_sign.png', 'stop_sign.png')

git_processor = AutoProcessor.from_pretrained("microsoft/git-base-coco")
git_model = AutoModelForCausalLM.from_pretrained("microsoft/git-base-coco")

blip_processor = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
blip_model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")

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.to(device)
blip_model.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_ids = 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 = generate_caption(git_processor, git_model, image)

    caption_blip = generate_caption(blip_processor, blip_model, image)

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

    return caption_git, caption_blip, caption_vitgpt

   
examples = [["cats.jpg"], ["stop_sign.png"]]

title = "Interactive demo: comparing image captioning models"
description = "Gradio Demo to compare GIT, BLIP and ViT-2-GPT2, 3 state-of-the-art captioning models. To use it, simply upload your image and click 'submit', or click one of the examples to load them. Read more at the links below."
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2102.03334' target='_blank'>ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision</a> | <a href='https://github.com/dandelin/ViLT' target='_blank'>Github Repo</a></p>"

interface = gr.Interface(fn=generate_captions, 
                         inputs=gr.inputs.Image(type="pil"),
                         outputs=[gr.outputs.Textbox(label="Caption generated by GIT"), gr.outputs.Textbox(label="Caption generated by BLIP"), gr.outputs.Textbox(label="Caption generated by ViT+GPT-2")], 
                         examples=examples, 
                         title=title,
                         description=description,
                         article=article, 
                         enable_queue=True)
interface.launch(debug=True)