handler.py

from typing import Dict, Any
import os
import requests
from io import BytesIO
from PIL import Image
import torch
from torchvision import transforms
from transformers import AutoModelForImageSegmentation

# Configuración
torch.set_float32_matmul_precision(["high", "highest"][0])
device = "cuda" if torch.cuda.is_available() else "cpu"

class EndpointHandler():
    def __init__(self, path=''):
        # Cargamos el modelo BiRefNet original (Efectivo y rápido)
        self.model = AutoModelForImageSegmentation.from_pretrained(
            'zhengpeng7/BiRefNet', 
            trust_remote_code=True
        )
        self.model.to(device)
        self.model.eval()
        self.model.half()

    def __call__(self, data: Dict[str, Any]):
        # 1. RECIBIR IMAGEN (Entrada Blindada)
        image_src = data["inputs"]
        image = None
        
        if isinstance(image_src, Image.Image):
            image = image_src
        elif isinstance(image_src, str):
            if image_src.startswith('http'):
                image = Image.open(BytesIO(requests.get(image_src).content))
            else:
                image = Image.open(image_src)
        else:
            image = Image.open(BytesIO(image_src))

        # 2. LIMPIEZA: Aseguramos RGB (Color Real)
        image = image.convert("RGB")
        orig_size = image.size
        
        # 3. PROCESAMIENTO IA
        transform = transforms.Compose([
            transforms.Resize((1024, 1024)),
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
        ])
        
        input_tensor = transform(image).unsqueeze(0).to(device).half()
        
        with torch.no_grad():
            preds = self.model(input_tensor)[-1].sigmoid().cpu()
        
        # 4. MÁSCARA
        pred = preds[0].squeeze()
        mask_pil = transforms.ToPILImage()(pred)
        mask_pil = mask_pil.resize(orig_size, resample=Image.Resampling.LANCZOS)
        
        # 5. APLICACIÓN FINAL (Sin tocar colores)
        image.putalpha(mask_pil)
        
        return image