# import gradio as gr
# import streamlit as st
# import torch 
# import re 
# from transformers import AutoTokenizer, ViTFeatureExtractor, VisionEncoderDecoderModel 

# device='cpu'
# encoder_checkpoint = "ydshieh/vit-gpt2-coco-en"
# decoder_checkpoint = "ydshieh/vit-gpt2-coco-en"
# model_checkpoint = "ydshieh/vit-gpt2-coco-eng"
# feature_extractor = ViTFeatureExtractor.from_pretrained(encoder_checkpoint)
# tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint)
# model = VisionEncoderDecoderModel.from_pretrained(model_checkpoint).to(device)

# def predict(image,max_length=64, num_beams=4):
#     input_image = Image.open(image)
#     model.eval()
#     pixel_values = feature_extractor(images=[input_image], return_tensors="pt").pixel_values
#     with torch.no_grad():
#         output_ids = model.generate(pixel_values, max_length=16, num_beams=4, return_dict_in_generate=True).sequences
#     preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
#     preds = [pred.strip() for pred in preds]  
#     return preds[0]
    
#   # image = image.convert('RGB')
#   # image = feature_extractor(image, return_tensors="pt").pixel_values.to(device)
#   # clean_text = lambda x: x.replace('<|endoftext|>','').split('\n')[0]
#   # caption_ids = model.generate(image, max_length = max_length)[0]
#   # caption_text = clean_text(tokenizer.decode(caption_ids))
#   # return caption_text 

# # st.title("Image to Text using Lora")

# inputs = gr.inputs.Image(label="Upload any Image", type = 'pil', optional=True)
# output = gr.outputs.Textbox(type="text",label="Captions")
# description = "NTT Data Bilbao team"
# title = "Image to Text using Lora"

# interface = gr.Interface(
#         fn=predict,
#         description=description,
#         inputs = inputs,
#         theme="grass",
#         outputs=output,
#         title=title,
#     )
# interface.launch(debug=True)

import torch 
import re 
import gradio as gr
from transformers import AutoTokenizer, ViTFeatureExtractor, VisionEncoderDecoderModel 

device='cpu'
encoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
decoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
model_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
print("------------------------- 1 -------------------------\n")
feature_extractor = ViTFeatureExtractor.from_pretrained(encoder_checkpoint)
print("------------------------- 2 -------------------------\n")
tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint
print("------------------------- 3 -------------------------\n")
model = VisionEncoderDecoderModel.from_pretrained(model_checkpoint).to(device)
print("------------------------- 4 -------------------------\n")


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


print("------------------------- 5 -------------------------\n")
input = gr.inputs.Image(label="Upload any Image", type = 'pil', optional=True)
output = gr.outputs.Textbox(type="auto",label="Captions")
examples = [f"example{i}.jpg" for i in range(1,7)]
print("------------------------- 6 -------------------------\n")
title = "Image Captioning "
description = "NTT Data"
interface = gr.Interface(
            
        fn=predict,
        description=description,
        inputs = input,
        theme="grass",
        outputs=output,
        examples = examples,
        title=title,
    )
interface.launch(debug=True)