Function sample_regions_camera_rays_and_features

plib/utils.py:1677–1758

Used in PCD version of inverse rendering. Args: cam_poses: intrinsics: sample_center: region_width_px: region_height_px: features: device: Returns:

(
        cam_poses: torch.Tensor,  # (m, 4, 4) target camera pose, cam to world
        intrinsics: torch.Tensor,  # (m, 3, 3)  intrinsic matrix of the camera
        sample_center: torch.Tensor,
        region_width_px: int,
        region_height_px: int,
        features: T.Dict[str, torch.Tensor] = None,
        device=torch.device('cpu'),
)

def sample_regions_camera_rays_and_features(
        cam_poses: torch.Tensor,  # (m, 4, 4) target camera pose, cam to world
        intrinsics: torch.Tensor,  # (m, 3, 3)  intrinsic matrix of the camera
        sample_center: torch.Tensor,
        region_width_px: int,
        region_height_px: int,
        features: T.Dict[str, torch.Tensor] = None,
        device=torch.device('cpu'),
):
    """
    Used in PCD version of inverse rendering.

    Args:
        cam_poses:
        intrinsics:
        sample_center:
        region_width_px:
        region_height_px:
        features:
        device:

    Returns:

    """
    uv_c = pinhole_projection(
        xyz_w=sample_center,
        intrinsics=intrinsics,
        H_c2w=cam_poses,
    )
    uv_c = torch.squeeze(uv_c)  # (m, 2)

    uv_offset = uv_c - torch.Tensor([region_width_px / 2, region_height_px / 2]).to(device=device).unsqueeze(0)

    ray_origins_w, ray_directions_w = generate_camera_rays(
        cam_poses=cam_poses,
        intrinsics=intrinsics,
        width_px=region_width_px,
        height_px=region_height_px,
        subsample=1,
        offsets=uv_offset,
        device=device,
    )  # (b*m, h, w, 3), (b*m, h, w, 3) normalized

    u, v = torch.meshgrid(
        torch.arange(0, region_width_px, 1, device=device),
        torch.arange(0, region_height_px, 1, device=device),
        indexing='xy',
    )  # u: (h', w') for x,  v: (h', w') for y in the sensor coord
    u = u + 0.5
    v = v + 0.5

    if features is not None:
        resampled_features = dict()
        for key in features.keys():
            feature = features[key]
            assert len(feature.shape) >= 3, "dimension of feature should contain at least n, h, w "
            if len(feature.shape) == 3:
                feature = feature.unsqueeze(-1)