main.py

import re
import torch
from fastapi import FastAPI, File, UploadFile
from fastapi.middleware.cors import CORSMiddleware
from transformers import DonutProcessor, VisionEncoderDecoderModel
from PIL import Image
from io import BytesIO

origins = [
    "https://duithive.vercel.app",
    "https://staging-duithive.vercel.app",
    "http://localhost:3000",
]

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=origins,
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2", use_fast=False)

model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")

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

task_prompt = "<s_cord-v2>"
decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids

# def generateOutput(fileData):
#     pil_image = Image.open(BytesIO(fileData))
#     resized_image = pil_image.resize((800, 600)).convert('RGB')
#     rgb_image = Image.new('RGB', resized_image.size)
#     rgb_image.paste(resized_image)
#     output_buffer = BytesIO()
#     rgb_image.save(output_buffer, format="JPEG", quality = 100)
#     jpeg_image = Image.open(BytesIO(output_buffer.getvalue()))
    
#     pixel_values = processor(jpeg_image, return_tensors="pt").pixel_values
#     outputs = model.generate(
#         pixel_values.to(device),
#         decoder_input_ids=decoder_input_ids.to(device),
#         max_length=model.decoder.config.max_position_embeddings,
#         pad_token_id=processor.tokenizer.pad_token_id,
#         eos_token_id=processor.tokenizer.eos_token_id,
#         use_cache=True,
#         bad_words_ids=[[processor.tokenizer.unk_token_id]],
#         return_dict_in_generate=True,
#     )
#     return outputs

def generateOutput(fileData):
    pil_image = Image.open(BytesIO(fileData))
    pil_image.resize((800, 600))
    pixel_values = processor(pil_image, return_tensors="pt").pixel_values

    outputs = model.generate(
        pixel_values.to(device),
        decoder_input_ids=decoder_input_ids.to(device),
        max_length=model.decoder.config.max_position_embeddings,
        pad_token_id=processor.tokenizer.pad_token_id,
        eos_token_id=processor.tokenizer.eos_token_id,
        use_cache=True,
        bad_words_ids=[[processor.tokenizer.unk_token_id]],
        return_dict_in_generate=True,
    )
    return outputs

@app.post("/ocr/")
async def analyze_image(file: UploadFile = File(...)):
    content = await file.read()
    outputs = generateOutput(content)
    sequence = processor.batch_decode(outputs.sequences)[0]
    sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
    sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
    return processor.token2json(sequence)