app.py

import gradio as gr
import torch
import numpy as np
import torchvision.transforms as T
from PIL import Image
from huggingface_hub import hf_hub_download

# Descargar modelo desde HuggingFace Hub (igual que Práctica 2 carga desde el Hub)
model_repo_id = "daniihc16/unet-grape-segmentation"

try:
    model_path = hf_hub_download(repo_id=model_repo_id, filename="unet_multiclass_hf.pth")
except Exception as e:
    print(f"Error descargando modelo de HuggingFace: {e}. Asegúrate de subir el modelo primero.")
    raise e

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = torch.jit.load(model_path, map_location=device)
model.eval()

# Clases y colores
class_names = ['background', 'leaves', 'wood', 'pole', 'grape']
class_colors = {
    0: [0, 0, 0],        # background - negro
    1: [0, 128, 0],      # leaves - verde
    2: [139, 69, 19],    # wood - marrón
    3: [128, 128, 128],  # pole - gris
    4: [128, 0, 128],    # grape - morado
}

def predict(image):
    if image is None:
        return None

    # Preprocesar igual que durante el entrenamiento
    transform = T.Compose([
        T.Resize((480, 640)),
        T.ToTensor(),
        T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ])
    tensor = transform(image).unsqueeze(0).to(device)

    # Inferencia
    with torch.no_grad():
        output = model(tensor)
    pred = torch.argmax(output, 1).cpu().numpy().reshape(480, 640)

    # Convertir máscara de clases a imagen RGB coloreada
    rgb = np.zeros((480, 640, 3), dtype=np.uint8)
    for cls_id, color in class_colors.items():
        rgb[pred == cls_id] = color

    return Image.fromarray(rgb)

iface = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil", label="Imagen de viñedo"),
    outputs=gr.Image(type="pil", label="Segmentación (background, leaves, wood, pole, grape)"),
    title="Segmentación Semántica de Viñedos",
    description=(
        "Modelo U-Net con backbone ResNet-50 entrenado para segmentar imágenes de viñedos. "
        "Clases: background (negro), leaves/hojas (verde), wood/madera (marrón), "
        "pole/poste (gris), grape/uva (morado). "
        f"Modelo cargado desde: daniihc16/unet-grape-segmentation"
    ),
    examples=['color_157.jpg', 'color_158.jpg']
)

if __name__ == "__main__":
    iface.launch()