annotator/dsine_hub.py

import torch
import numpy as np
from PIL import Image


class NormalDetector:
    def __init__(self):
        self.model_path = "hugoycj/DSINE-hub"
        self.dsine = torch.hub.load(self.model_path, "DSINE", trust_repo=True)
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    @torch.no_grad()
    def __call__(self, image):
        self.dsine.model.to(self.device)
        self.dsine.model.pixel_coords = self.dsine.model.pixel_coords.to(self.device)
        H, W, C = image.shape

        normal = self.dsine.infer_pil(image)[0]  # Output shape: (H, W, 3)
        normal = (normal + 1.0) / 2.0  # Convert values to the range [0, 1]
        normal = (normal * 255).cpu().numpy().astype(np.uint8).transpose(1, 2, 0)
        normal_img = Image.fromarray(normal).resize((W, H))

        self.dsine.model.to("cpu")
        self.dsine.model.pixel_coords = self.dsine.model.pixel_coords.to("cpu")
        return normal_img


if __name__ == "__main__":
    from diffusers.utils import load_image

    image = load_image(
        "https://qhstaticssl.kujiale.com/image/jpeg/1716177580588/9AAA49344B9CE33512C4EBD0A287495F.jpg"
    )
    image = np.asarray(image)
    normal_detector = NormalDetector()
    normal_image = normal_detector(image)
    normal_image.save("normal_image.jpg")