[TensorFlow] Official implementation of CVPR'20 oral paper - D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features https://arxiv.org/abs/2003.03164
Hi, I appreciate your great work!
I am currently trying to register a point cloud frame from RGBD to a whole indoor map using data from the Redwood dataset. Since the map is much larger than one frame point cloud, I have sliced the map into several submaps. When I split the map into 27 pieces and generate descriptors with test_3dMatch.py, it reports an OOM error. However, it works well when I split the same map into 64 pieces, meaning that each submap is smaller. The whole map contains 13 million points before downsampling it with voxel_down_sample(voxel_size=0.03), and each submap has overlap with the neighboring submap.
I am wondering what the probable reason for the OOM error is. I tested it on my 2080Ti GPU with 8GB of capable space.
Thanks a lot!
Hi, I appreciate your great work! I am currently trying to register a point cloud frame from RGBD to a whole indoor map using data from the Redwood dataset. Since the map is much larger than one frame point cloud, I have sliced the map into several submaps. When I split the map into 27 pieces and generate descriptors with test_3dMatch.py, it reports an OOM error. However, it works well when I split the same map into 64 pieces, meaning that each submap is smaller. The whole map contains 13 million points before downsampling it with voxel_down_sample(voxel_size=0.03), and each submap has overlap with the neighboring submap. I am wondering what the probable reason for the OOM error is. I tested it on my 2080Ti GPU with 8GB of capable space. Thanks a lot!