SiamVFE & SiamGCN-GCA

This is the official implementation of SiamVFE and SiamGCN-GCA for the task of point cloud change detection for city scenes.

Install

1. Create an anaconda environment with python 3.8.5 conda create -n "siam" python=3.8.5

2. Activate the anaconda environment conda activate siam

3. Install the pytorch version, given your CUDA setup. i.e. for pytorch 1.10.1 and CUDA 11.3 conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge

4. Install pytorch-geometric conda install -c pyg pyg

5. Install laspy pip install laspy[lazrs,laszip]

6. Install imbalanced-learn conda install -c conda-forge imbalanced-learn

7. Install matplotlib conda install -c conda-forge matplotlib

Supported Datasets

• SHREC2023

SHREC2023

Place the SHREC2023 dataset folder "PCLChange", in the main folder. The structure should be like:

• SiamVFE/PCLChange/
  • lidar/
    • 2016/
    • 2020/
    • labeled_point_lists_train/
  • synthetic_city_scenes
    • time_a/
    • time_b/
    • labeled_point_lists_train_syn/

Training

SiamVFE

• on lidar data python main_siamvfe.py --batch_size 32 --num_workers 4 --data lidar

• on synthetic data python main_siamvfe.py --batch_size 32 --num_workers 4 --data synthetic

SiamGCN-GCA

• on lidar data python main_siamgcn_gca.py --batch_size 16 --num_workers 4 --data lidar

• on synthetic data python main_siamgcn_gca.py --batch_size 16 --num_workers 4 --data synthetic

If out of GPU memory, reduce the batch size.

Pre-processing

When executing either training script for the first time, the pre-processing step will take place. This is performed only once, to extract the scenes of interest and construct the dataset.

Evaluate models

To evaluate a model run the training script with the --test_model flag set to True.

Also, select the type of data, lidar or synthetic

i.e. python main_siamgcn_gca.py --batch_size 4 --num_workers 4 --test_model True --data lidar

Acknowledgements

This work relies upon code from OpenPCDet and SiamGCN.