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')
torch.hub.download_url_to_file('https://cdn.openai.com/dall-e-2/demos/text2im/astronaut/horse/photo/0.jpg', 'astronaut.jpg')

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

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

git_model.to(device)

def generate_caption(processor, model, image):
    inputs = processor(images=image, return_tensors="pt").to(device)
    generated_ids = model.generate(pixel_values=inputs.pixel_values, max_length=50)
    generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
   
    return generated_caption


def generate_captions(image):
    return generate_caption(git_processor, git_model, image)

   
examples = [["cats.jpg"], ["stop_sign.png"], ["astronaut.jpg"]]
outputs = [gr.outputs.Textbox(label="Caption generated by GIT-Large coco")]

title = "Interactive demo: GIT-Large coco image captioning"
description = "GIT-Large coco image captioning"
article = "<p style='text-align: center'><a href='https://huggingface.co/docs/transformers/main/model_doc/blip' target='_blank'>BLIP docs</a> | <a href='https://huggingface.co/docs/transformers/main/model_doc/git' target='_blank'>GIT docs</a></p>"

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