Closed Steven-ZY closed 3 years ago
Hi Steven, Siléane dataset is encoded in 2.5D format: 3D coordinates can be derived for each pixel of the depth maps, and point-wise instance segmentation can be derived from the segmentation maps. See https://rbregier.github.io/dataset2017 for more information.
You can easily convert it to an unorganized 3D point cloud representation with point-wise label if you prefer (e.g. taking inspiration from https://github.com/rbregier/pose_recovery_evaluation/blob/c726dff99beb320cf3d41348a36362c573c142cd/visualization.py#L48 ), but we do not provide any tools to do so however.
Hi, we can get the segmentation GT described as a PNG. But, is there any point-wise label file just like ShapeNet Dataset?