Self-supervised multi-echo point cloud denoising in snowfall, Publication

Citation:

@article{seppanen2024self, title={Self-supervised multi-echo point cloud denoising in snowfall}, author={Sepp{"a}nen, Alvari and Ojala, Risto and Tammi, Kari}, journal={Pattern Recognition Letters}, year={2024}, publisher={Elsevier} }

Environment:

• Python 3.8.10
• CUDA 11.6
• PyTorch 1.12.1+cu116
• Numpy 1.23.3

Datasets:

• Download SnowyKITTI
• Download STF

Collect corrupted point clouds from STF:

cd utils
python3 create_new_stf.py -d root/STF_dataset/ -n root_for_new_dataset/

Train:

cd networks
./self_train.sh -d root/snowyKITTI/dataset/ -a smednet.yml -l /your/log/folder/ -c 0
./multi_self_train.sh -d root/new_STF/dataset/ -a smednet.yml -l /your/log/folder/ -c 0 

Infer (pretrained singe-echo model -m root/logs/2023-2-21-15:49/, multi-echo model -m root/logs/2023-2-27-13:11/):

cd networks/train/tasks/semantic
python3 self_infer.py -d root/snowyKITTI/dataset/ -m root/logs/2023-2-21-15:49/ -l /your/predictions/folder/ -s test
python3 multi_self_infer.py -d root/new_STF/dataset/ -m root/logs/2023-2-27-13:11/ -l /your/predictions/folder/ -s test
(-s = split)

Evaluate:

cd networks/train/tasks/semantic
python3 evaluate_iou.py -d root/snowyKITTI/dataset/ -dc root/networks/train/tasks/semantic/config/labels/snowy-kitti.yaml -p /your/predictions/folder/ -s test
(-s = split)

Visualize:

cd utils
single-echo:
python3 visualize.py -d root/snowyKITTI/dataset/ -c root/networks/train/tasks/semantic/config/labels/snowy-kitti.yaml -p /your/predictions/folder/ -s 22
multi-echo:
python3 visualize.py -d root/new_STF/dataset/ -c root/networks/train/tasks/semantic/config/labels/stf.yaml -p /your/predictions/folder/ -s 4 -me 
(-me = multi-echo)