KernelA/pytorch-point-transformer

Implementation of point transformer for point cloud classification and segmentation

Pytorch Point Transformer

Original article

Results

Modelnet 40

WandDB project (can be unavailable in the future)

WandDB report (can be unavailable in the future)

You can download a trained model from the model registry.

Description

Presentation

Colab training

More info

Requirements

1. Python 3.8 or higher.
2. CUDA 11.3 or higher

How to run

CPU (training will be slow)

pip install -r ./requirements.torch.gpu.txt
pip install -r ./requirements.cpu.txt -r ./requirements.base.txt

GPU:

pip install -r ./requirements.torch.gpu.txt
pip install -r ./requirements.gpu.txt -r ./requirements.base.txt

Only for presentation.ipynb

pip install -r requirements.add.txt

Train on the simple shapes dataset

dvc repro pipelines/simple_shapes/dvc.yaml:train

Train on the ModelNet10/40 dataset

dvc repro pipelines/modelnet10/dvc.yaml:train

or

dvc repro pipelines/modelnet40/dvc.yaml:train

Train segmentation model on PartNet

1. Download dataset here
2. Run: python ./preprocess_partnet.py --data_dir <data_dir> --out_dir <where_to_store_processed>
3. Edit: partnet.yaml. Set dataset_dir to the <where_to_store_processed>
4. Run: dvc repro pipelines/partnet/dvc.yaml:train