from PIL import Image
import requests
import gradio as gr

from transformers import BlipProcessor, BlipForConditionalGeneration



model = BlipForConditionalGeneration.from_pretrained('jaimin/Imagecap')
processor = BlipProcessor.from_pretrained('jaimin/Imagecap')



def predict(image,max_length=64, num_beams=4):
  image = image.convert('RGB')
  #image = feature_extractor(image, return_tensors="pt").pixel_values.to(device)
  inputs = processor(image, return_tensors="pt")
  #clean_text = lambda x: x.replace('<|endoftext|>','').split('\n')[0]
  caption_ids = model.generate(inputs, max_length = max_length)[0]
  caption_text = tokenizer.decode(caption_ids)
  return processor.decode(caption_ids[0], skip_special_tokens=True)



input = gr.inputs.Image(label="Upload your Image", type = 'pil', optional=True)
output = gr.outputs.Textbox(label="Captions")

title = "ImageCap"

interface = gr.Interface(
        fn=predict,
        inputs = input,
        outputs=output,
        title=title,
    )
interface.launch(debug=True)