Tiktok-Talent-Info/endpoints/image.py

from fastapi import UploadFile, Form
from fastapi.responses import JSONResponse
import io
import base64
import asyncio
import numpy as np
from PIL import Image
from pipeline_setup import pipe, IMAGE_TOKEN
from utils.image_processing import encode_image_base64

# async def image_query(file: UploadFile, question: str = Form(...)):
#     """
#     API endpoint to process an image with the user's query.
#     """
#     try:
#         if file.content_type not in ["image/jpeg", "image/png"]:
#             return JSONResponse({"query": question, "error": "Unsupported file type."})

#         image_data = await file.read()
#         image = Image.open(io.BytesIO(image_data)).convert("RGB").resize((512, 512))
#         encoded_image_base64 = encode_image_base64(image)
        
#         question_with_image_token = f"{question}\n{IMAGE_TOKEN}"
#         response = await asyncio.to_thread(pipe, (question, image))
#         return JSONResponse({"query": question, "response": response.text})
#     except Exception as e:
#         return JSONResponse({"query": question, "error": str(e)})


# import mimetypes
async def image_query(image: np.ndarray, question: str):
    """
    API endpoint to process an image (as numpy array) with the user's query.
    """
    try:
        # Convert the numpy array to a PIL Image
        image = Image.fromarray(image).convert("RGB").resize((512, 512))
        
        # Encode the image to base64 (optional, if needed by your pipeline)
        buffered = io.BytesIO()
        image.save(buffered, format="JPEG")
        encoded_image_base64 = base64.b64encode(buffered.getvalue()).decode("utf-8")

        # Prepare the query with the image token
        question_with_image_token = f"{question}\n{IMAGE_TOKEN}"

        # Query the model
        response = await asyncio.to_thread(pipe, (question, image))
        return {"query": question, "response": response.text}
    except Exception as e:
        return {"query": question, "error": str(e)}