INNER CODE UNIT · Python

ori_imgs

junshutang/Make-It-3D · main.py:185

    ori_imgs = ref_imgs[:, :3, :, :] * ref_imgs[:, 3:, :, :] + (1 - ref_imgs[:, 3:, :, :])
    
    mask = imgs[:, :, 3:]
    # mask[mask < 0.5 * 255] = 0
    # mask[mask >= 0.5 * 255] = 1 
    kernel = np.ones(((5,5)), np.uint8) ##11
    mask = cv2.erode(mask,kernel,iterations=1)
    mask = (mask == 0)
    mask = (torch.from_numpy(mask)).unsqueeze(0).unsqueeze(0).to(device)
    depth_mask = mask
    
    # depth estimation
    with torch.no_grad():
        depth_prediction = depth_model.forward(depth_transform(ori_imgs))
        depth_prediction = torch.nn.functional.interpolate(
            depth_prediction.unsqueeze(1),
            size=512,
            mode="bicubic",

View source record →