search

Pointcept / PointTransformerV3

[CVPR'24 Oral] Official repository of Point Transformer V3 (PTv3)
MIT License
583 stars 30 forks source link

readme

Point Transformer V3

PWC
PWC
PWC
PWC
PWC

This repo is the official project repository of the paper Point Transformer V3: Simpler, Faster, Stronger and is mainly used for releasing schedules, updating instructions, sharing experiment records (containing model weight), and handling issues. The code will be updated in Pointcept v1.5.
[ Backbone ] [PTv3] - [ arXiv ] [ Bib ] [ Code ]

teaser

Highlights

Overview

Schedule

To make our polished code and reproduced experiments available as soon as possible, this time we will release what we already finished immediately after a validation instead of releasing them together after all work is done. We list a task list as follows:

Citation

If you find PTv3 useful to your research, please cite our work as an acknowledgment. (ΰ©­ΛŠκ’³β€‹Λ‹)੭✧

@inproceedings{wu2024ptv3,
    title={Point Transformer V3: Simpler, Faster, Stronger},
    author={Wu, Xiaoyang and Jiang, Li and Wang, Peng-Shuai and Liu, Zhijian and Liu, Xihui and Qiao, Yu and Ouyang, Wanli and He, Tong and Zhao, Hengshuang},
    booktitle={CVPR},
    year={2024}
}

@inproceedings{wu2024ppt,
    title={Towards Large-scale 3D Representation Learning with Multi-dataset Point Prompt Training},
    author={Wu, Xiaoyang and Tian, Zhuotao and Wen, Xin and Peng, Bohao and Liu, Xihui and Yu, Kaicheng and Zhao, Hengshuang},
    booktitle={CVPR},
    year={2024}
}

@inproceedings{wu2022ptv2,
    title={Point transformer V2: Grouped Vector Attention and Partition-based Pooling},
    author={Wu, Xiaoyang and Lao, Yixing and Jiang, Li and Liu, Xihui and Zhao, Hengshuang},
    booktitle={NeurIPS},
    year={2022}
}

@misc{pointcept2023,
    title={Pointcept: A Codebase for Point Cloud Perception Research},
    author={Pointcept Contributors},
    howpublished={\url{https://github.com/Pointcept/Pointcept}},
    year={2023}
}

Installation

Requirements

PTv3 relies on FlashAttention, while FlashAttention relies on the following requirement, make sure your local Pointcept environment satisfies the requirements:

(Recommendation)

If you can not upgrade your local environment to satisfy the above-recommended requirements, the following requirement is the minimum to run PTv3 with Pointcept, and you need to disable Flash Attention to enable PTv3:

(Minimum)

Environment

cd libs/pointops python setup.py install cd ../..

spconv (SparseUNet)

refer https://github.com/traveller59/spconv

pip install spconv-cu118 # choose version match your local cuda version

Open3D (visualization, optional)

pip install open3d


- Flash Attention

Following [README](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#installation-and-features) in Flash Attention repo and install Flash Attention for PTv3. This installation is optional, but we recommend enabling Flash Attention for PTv3.

## Data Preparation
Please further refer Pointcept readme [Data Preparation](https://github.com/Pointcept/Pointcept#data-preparation) section.

## Quick Start
### Two running scenarios
We provide two running scenarios for PTv3, Pointcept-driven and custom-framework-driven. For the former one, you only need to clone the code of Pointcept to your local and follow the [Quick Start](https://github.com/Pointcept/Pointcept#quick-start) in Pointcept to run PTv3:
```bash
git clone https://github.com/Pointcept/Pointcept.git
sh scripts/train.sh -p ${INTERPRETER_PATH} -g ${NUM_GPU} -d ${DATASET_NAME} -c ${CONFIG_NAME} -n ${EXP_NAME}

For the latter scenario, we offer a distinct instance of PTv3, disassociated from our Pointcept framework. To incorporate this code into your project, clone the project repo and copy the following file/folder to your project:

git clone https://github.com/Pointcept/PointTransformerV3.git
cp model.py ${PATH_TO_YOUR_PROJECT}
cp -r serialization ${PATH_TO_YOUR_PROJECT}

Align the input dictionary defined in our model file and the model will return the encoded feature of the given batch point cloud.

Flash Attention

The full PTv3 relies on Flash Attention, while Flash Attention relies on CUDA 11.6 and above, make sure your local Pointcept environment satisfies the requirements.

If you can not upgrade your local environment to satisfy the requirements (CUDA >= 11.6), then you can disable FlashAttention by setting the model parameter enable_flash to false and reducing the enc_patch_size and dec_patch_size to a level (e.g. 128).

FlashAttention force disables RPE and forces the accuracy reduced to fp16. If you require these features, please disable enable_flash and adjust enable_rpe, upcast_attention andupcast_softmax.

Model Zoo

1. Indoor semantic segmentation

Model Benchmark Additional Data Num GPUs Val mIoU Config Tensorboard Exp Record
PTv3 ScanNet ✗ 4 77.6% link link link
PTv3 + PPT ScanNet ✓ 8 78.5% link link link
PTv3 ScanNet200 ✗ 4 35.3% link link link
PTv3 + PPT ScanNet200 ✓ (f.t.) 4
PTv3 S3DIS (Area5) ✗ 4 73.6% link link link
PTv3 + PPT S3DIS (Area5) ✓ 8 75.4% link link link

*Released model weights are temporarily invalid as the model structure of PTv3 is adjusted.

Example running scripts are as follows:

# Scratched ScanNet
sh scripts/train.sh -g 4 -d scannet -c semseg-pt-v3m1-0-base -n semseg-pt-v3m1-0-base
# PPT joint training (ScanNet + Structured3D) and evaluate in ScanNet
sh scripts/train.sh -g 8 -d scannet -c semseg-pt-v3m1-1-ppt-extreme -n semseg-pt-v3m1-1-ppt-extreme

# Scratched ScanNet200
sh scripts/train.sh -g 4 -d scannet200 -c semseg-pt-v3m1-0-base -n semseg-pt-v3m1-0-base
# Fine-tuning from  PPT joint training (ScanNet + Structured3D) with ScanNet200
# TODO

# Scratched S3DIS, S3DIS rely on RPE, also an example for disable flash attention
sh scripts/train.sh -g 4 -d s3dis -c semseg-pt-v3m1-0-rpe -n semseg-pt-v3m1-0-rpe
# PPT joint training (ScanNet + S3DIS + Structured3D) and evaluate in ScanNet
sh scripts/train.sh -g 8 -d s3dis -c semseg-pt-v3m1-1-ppt-extreme -n semseg-pt-v3m1-1-ppt-extreme

# More configs and exp records for PTv3 will be available soon.

2.Outdoor semantic segmentation

Model Benchmark Additional Data Num GPUs Val mIoU Config Tensorboard Exp Record
PTv3 nuScenes ✗ 4 80.3 link link link
PTv3 + PPT nuScenes ✓ 8
PTv3 SemanticKITTI ✗ 4
PTv3 + PPT SemanticKITTI ✓ 8
PTv3 Waymo ✗ 4 71.2 link link link (log only)
PTv3 + PPT Waymo ✓ 8

*Released model weights are temporarily invalid as the model structure of PTv3 is adjusted.
*Model weights trained with Waymo Open Dataset cannot be released due to the regulations.

Example running scripts are as follows:

# Scratched ScanNet
sh scripts/train.sh -g 4 -d scannet -c semseg-pt-v3m1-0-base -n semseg-pt-v3m1-0-base
# PPT joint training (ScanNet + Structured3D) and evaluate in ScanNet
sh scripts/train.sh -g 8 -d scannet -c semseg-pt-v3m1-1-ppt-extreme -n semseg-pt-v3m1-1-ppt-extreme

# Scratched ScanNet200
sh scripts/train.sh -g 4 -d scannet200 -c semseg-pt-v3m1-0-base -n semseg-pt-v3m1-0-base
# Fine-tuning from  PPT joint training (ScanNet + Structured3D) with ScanNet200
# TODO

# Scratched S3DIS, S3DIS rely on RPE, also an example for disable flash attention
sh scripts/train.sh -g 4 -d s3dis -c semseg-pt-v3m1-0-rpe -n semseg-pt-v3m1-0-rpe
# PPT joint training (ScanNet + S3DIS + Structured3D) and evaluate in ScanNet
sh scripts/train.sh -g 8 -d s3dis -c semseg-pt-v3m1-1-ppt-extreme -n semseg-pt-v3m1-1-ppt-extreme
# S3DIS 6-fold cross validation
# 1. The default configs are evaluated on Area_5, modify the "data.train.split", "data.val.split", and "data.test.split" to make the config evaluated on Area_1 ~ Area_6 respectively.
# 2. Train and evaluate the model on each split of areas and gather result files located in "exp/s3dis/EXP_NAME/result/Area_x.pth" in one single folder, noted as RECORD_FOLDER.
# 3. Run the following script to get S3DIS 6-fold cross validation performance:
export PYTHONPATH=./
python tools/test_s3dis_6fold.py --record_root ${RECORD_FOLDER}

# Scratched nuScenes
sh scripts/train.sh -g 4 -d nuscenes -c semseg-pt-v3m1-0-base -n semseg-pt-v3m1-0-base
# Scratched Waymo
sh scripts/train.sh -g 4 -d waymo -c semseg-pt-v3m1-0-base -n semseg-pt-v3m1-0-base

# More configs and exp records for PTv3 will be available soon.