Open
Description
in kitti_sequence.py you code the mask as this and what the instance's rays mean? Do you have any theory?
if max_num_matchess > pixels_coord.shape[0] * 0.5:
n = np.argmax(num_matches)
instance.mask = masks_2d[n, ...]
instance.bbox = bboxes_2d[n, ...]
if instance.mask[instance.mask].shape[0] > self.min_mask_area:
# Sample non-surface pixels
non_surface_pixels = self.pixels_sampler(instance.bbox, instance.mask)
if non_surface_pixels.shape[0] > 200:
sample_ind = np.linspace(0, non_surface_pixels.shape[0]-1, 200).astype(np.int32)
non_surface_pixels = non_surface_pixels[sample_ind, :]
pixels_inside_bb = np.concatenate([pixels_uv, non_surface_pixels], axis=0)
# rays contains all, but depth should only contain foreground
instance.rays = get_rays(pixels_inside_bb, self.invK).astype(np.float32)
instance.depth = surface_points[:, 2].astype(np.float32)
Metadata
Assignees
Labels
No labels