Point cloud geometry upsampling
A method, computer program, and computer system is provided for processing point cloud data. Quantized point cloud data including a plurality of voxels is received. An occupancy map is generated for the quantized point cloud corresponding to lost voxels during quantization from among the plurality of voxels. A point cloud is reconstructed from the quantized point cloud data based on populating the lost voxels.
1. A method of point cloud processing, executable by a processor, comprising:
receiving quantized point cloud data including a plurality of voxels;
generating an occupancy map for the quantized point cloud corresponding to lost voxels during quantization from among the plurality of voxels; and
reconstructing a point cloud from the quantized point cloud data based on populating the lost voxels.
2. The method of claim 1 , wherein the occupancy map corresponds to a prediction probability for each voxel from among the plurality of voxels being lost.
3. The method of claim 2 , wherein the lost voxels are populated based on the prediction probability being greater than a threshold value.
4. The method of claim 1 , wherein the point cloud is reconstructed based on dividing the quantized point cloud data into one or more patches and upsampling the patches.
5. The method of claim 1 , wherein the point cloud is reconstructed based on minimizing a binary cross-entropy classification loss associated with the occupancy map.
6. The method of claim 1 , wherein the point cloud is reconstructed through one or more from among a U-net, a spatial pyramid pooling, or a 3D convolution.
7. The method of claim 6 , wherein the 3D convolution comprises one or more from among a convolution, a submanifold convolution, a dilated convolution, and an atrous convolution.
8. A computer system for point cloud processing, the computer system comprising:
one or more computer-readable non-transitory storage media configured to store computer program code; and
one or more computer processors configured to access said computer program code and operate as instructed by said computer program code, said computer program code including:
receiving code configured to cause the one or more computer processors to receive quantized point cloud data including a plurality of voxels;
generating code configured to cause the one or more computer processors to generate an occupancy map for the quantized point cloud corresponding to lost voxels during quantization from among the plurality of voxels; and
reconstructing code configured to cause the one or more computer processors to reconstruct a point cloud from the quantized point cloud data based on populating the lost voxels.
9. The computer system of claim 8 , wherein the occupancy map corresponds to a prediction probability for each voxel from among the plurality of voxels being lost.
10. The computer system of claim 9 , wherein the lost voxels are populated based on the prediction probability being greater than a threshold value.
11. The computer system of claim 8 , wherein the point cloud is reconstructed based on dividing the quantized point cloud data into one or more patches and upsampling the patches.
12. The computer system of claim 8 , wherein the point cloud is reconstructed based on minimizing a binary cross-entropy classification loss associated with the occupancy map.
13. The computer system of claim 8 , wherein the point cloud is reconstructed through one or more from among a U-net, a spatial pyramid pooling, or a 3D convolution.
14. The computer system of claim 13 , wherein the 3D convolution comprises one or more from among a convolution, a submanifold convolution, a dilated convolution, and an atrous convolution.
15. A non-transitory computer readable medium having stored thereon a computer program for point cloud processing, the computer program configured to cause one or more computer processors to:
receive quantized point cloud data including a plurality of voxels;
generate an occupancy map for the quantized point cloud corresponding to lost voxels during quantization from among the plurality of voxels; and
reconstruct a point cloud from the quantized point cloud data based on populating the lost voxels.
16. The computer readable medium of claim 15 , wherein the occupancy map corresponds to a prediction probability for each voxel from among the plurality of voxels being lost.
17. The computer readable medium of claim 16 , wherein the lost voxels are populated based on the prediction probability being greater than a threshold value.
18. The computer readable medium of claim 15 , wherein the point cloud is reconstructed based on minimizing a binary cross-entropy classification loss associated with the occupancy map.
19. The computer readable medium of claim 15 , wherein the point cloud is reconstructed through one or more from among a U-net, a spatial pyramid pooling, or a 3D convolution.
20. The computer readable medium of claim 19 , wherein the 3D convolution comprises one or more from among a convolution, a submanifold convolution, a dilated convolution, and an atrous convolution.