https://github.com/NJUVISION/PCGCv2
这个是8.3下载的版本,较之前的模型不一样,We have simplified the code, and use torchac to replace tensorflow-compression for arithmetic coding in the updated version.
Multiscale Point Cloud Geometry Compression
We apply an end-to-end learning framework to compress the 3D point cloud geometry (PCG) efficiently. Leveraging the sparsity nature of point cloud, we introduce the multiscale structure to represent native PCG compactly, offering the hierarchical reconstruction capability via progressive learnt re-sampling. Under this framework, we devise the sparse convolution-based autoencoder for feature analysis and aggregation. At the bottleneck layer, geometric occupancy information is losslessly encoded with a very small percentage of bits consumption, and corresponding feature attributes are lossy compressed.
News
- 2021.1.1 Our paper has been accepted by DCC2021! [paper] [presentation]
- 2021.2.25 We have updated MinkowskiEngine to v0.5. The bug on GPU is fixed. And the encoding and decoding runtime is reduced.
- 2021.7.28 We have simplified the code, and use torchac to replace tensorflow-compression for arithmetic coding in the updated version. And the old version can be found here.
Requirments 云脑镜像: yeh:PCGCv2_04
- python3.7 or 3.8 云脑:3.6.9
- cuda10.2 or 11.0 云脑:V10.2.89
- pytorch1.7 or 1.8 云脑:1.8.1
- MinkowskiEngine 0.5 or higher (for sparse convolution) 云脑:0.5.3
- torchac 0.9.3 (for arithmetic coding) https://github.com/fab-jul/torchac 云脑:pip install torchac,这个是根据累积分布函数进行算术编解码的,并且集成到了pytorch
- tmc3 v12 (for lossless compression of downsampled point cloud coordinates) https://github.com/MPEGGroup/mpeg-pcc-tmc13 云脑:不能用旧版本,按照链接下载安装,然后把生成的tmc3文件拷到PCGCv2下
We recommend you to follow https://github.com/NVIDIA/MinkowskiEngine to setup the environment for sparse convolution.
Usage
Testing
Please download the pretrained models and install tmc3 mentioned above first.
sudo chmod 777 tmc3 pc_error_d
python coder.py --filedir='testdata/8iVFB/redandblack_vox10_1550.ply' --ckptdir='ckpts/r7_0.4bpp.pth'
python test.py --filedir='longdress_vox10_1300.ply'
The testing rusults of 8iVFB can be found in ./results
Training
python train.py --dataset='training_dataset_rootdir'
python train.py --dataset='../../training_dataset/' --init_ckpt='ckpts/r3_0.10bpp.pth' --epoch=3
Authors
These files are provided by Nanjing University Vision Lab. And thanks for the help from Prof. Dandan Ding from Hangzhou Normal University and Prof. Zhu Li from University of Missouri at Kansas. Please contact us (mazhan@nju.edu.cn and wangjq@smail.nju.edu.cn) if you have any questions.