Awesome

3D Semantic Segmentation in the Wild: Learning Generalized Models for Adverse-Condition Point Clouds

The official implementation of "3D Semantic Segmentation in the Wild: Learning Generalized Models for Adverse-Condition Point Clouds". (CVPR 2023) :fire::fire::fire:

:fire: For more information follow the PAPER link!:fire:

Authors: Aoran Xiao, Jiaxing Huang, Weihao Xuan, Ruijie Ren, Kangcheng Liu, Dayan Guan, Abdulmotaleb El Saddik, Shijian Lu, Eric Xing

SemanticSTF dataset

Download SemanticSTF dataset from GoogleDrive, BaiduYun(code: 6haz). Data folders are as follows: The data should be organized in the following format:

/SemanticSTF/
  └── train/
    └── velodyne
      └── 000000.bin
      ├── 000001.bin
      ...
    └── labels
      └── 000000.label
      ├── 000001.label
      ...
  └── val/
      ...
  └── test/
      ...
  ...
  └── semanticstf.yaml

We provide class annotations in 'semanticstf.yaml'

PointDR

Baseline code for 3D LiDAR Domain Generalization

cd pointDR/

Installation

GPU Requirement: > 1 x NVIDIA GeForce RTX 2080 Ti.

The code has been tested with

• Python 3.8, CUDA 10.2, Pytorch 1.8.0, TorchSparse 1.4.0.
• Python 3.8, CUDA 11.6, Pytorch 1.13.0, TorchSparse 2.0.0b0
• IMPORTANT: This code base is not compatible with TorchSparse 2.1.0.

Please refer to here for the installation details.

Pip/Venv/Conda

In your virtual environment follow TorchSparse. This will install all the base packages.

Data preparation

SynLiDAR

Download SynLiDAR dataset from here, then prepare data folders as follows:

./
├── 
├── ...
└── path_to_data_shown_in_config/
    └──sequences/
        ├── 00/           
        │   ├── velodyne/	
        |   |	├── 000000.bin
        |   |	├── 000001.bin
        |   |	└── ...
        │   └── labels/ 
        |       ├── 000000.label
        |       ├── 000001.label
        |       └── ...
        └── 12/

SemanticKITTI

To download SemanticKITTI follow the instructions here. Then, prepare the paths as follows:

./
├── 
├── ...
└── path_to_data_shown_in_config/
      └── sequences
            ├── 00/           
            │   ├── velodyne/	
            |   |	   ├── 000000.bin
            |   |	   ├── 000001.bin
            |   |	   └── ...
            │   ├── labels/ 
            |   |      ├── 000000.label
            |   |      ├── 000001.label
            |   |      └── ...
            |   ├── calib.txt
            |   ├── poses.txt
            |   └── times.txt
            └── 08/

• Don't forget revise the data root dir in configs/kitti2stf/default.yaml and configs/synlidar2stf/default.yaml

Training

For SemanticKITTI->SemanticSTF, run:

python train.py configs/kitti2stf/minkunet/cr0p5.yaml

For SynLiDAR->SemanticSTF, run:

python train.py configs/synlidar2stf/minkunet/cr0p5.yaml

Testing

For SemanticKITTI->SemanticSTF, run:

python evaluate.py configs/kitti2stf/minkunet/cr0p5.yaml --checkpoint_path /PATH/CHECKPOINT_NAME.pt

For SynLiDAR->SemanticSTF, run:

python evaluate_by_weather.py configs/synlidar2stf/minkunet/cr0p5.yaml  --checkpoint_path /PATH/CHECKPOINT_NAME.pt

You can download the pretrained models on both SemanticKITTI->SemanticSTF and SynLiDAR->SemanticSTF from here

TODO List

• Release of SemanticSTF dataset. :rocket:
• Release of code of PointDR. :rocket:
• Add license. See here for more details.
• Multi-modal UDA for normal-to-adverse weather 3DSS.

References

If you find our work useful in your research, please consider citing:

@article{xiao20233d,
  title={3D Semantic Segmentation in the Wild: Learning Generalized Models for Adverse-Condition Point Clouds},
  author={Xiao, Aoran and Huang, Jiaxing and Xuan, Weihao and Ren, Ruijie and Liu, Kangcheng and Guan, Dayan and Saddik, Abdulmotaleb El and Lu, Shijian and Xing, Eric},
  journal={arXiv preprint arXiv:2304.00690},
  year={2023}
}

SemanticSTF dataset consists of re-annotated LiDAR point cloud data from the STF dataset. Kindly consider citing it if you intend to use the data:

@inproceedings{bijelic2020seeing,
  title={Seeing through fog without seeing fog: Deep multimodal sensor fusion in unseen adverse weather},
  author={Bijelic, Mario and Gruber, Tobias and Mannan, Fahim and Kraus, Florian and Ritter, Werner and Dietmayer, Klaus and Heide, Felix},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={11682--11692},
  year={2020}
}

License

<a rel="license" href="http://creativecommons.org/licenses/by-nc-sa/4.0/"><img alt="Creative Commons License" style="border-width:0" src="https://i.creativecommons.org/l/by-nc-sa/4.0/80x15.png" /></a> <br /> This work is under the <a rel="license" href="http://creativecommons.org/licenses/by-nc-sa/4.0/">Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License</a>.

Recommended Repos

Check our other repos for point cloud understanding!

• Learning From Synthetic LiDAR Sequential Point Cloud for Semantic Segmentation (AAAI2022)
• PolarMix: A General Data Augmentation Technique for LiDAR Point Clouds (NeurIPS 2022)
• Unsupervised Point Cloud Representation Learning with Deep Neural Networks: A Survey (TPAMI2023)

Thanks

We thank the opensource projects TorchSparse, SPVNAS and SeeingThroughFog.