PointNetVlad-Pytorch

Unofficial PyTorch implementation of PointNetVlad (https://github.com/mikacuy/pointnetvlad) refer to https://github.com/cattaneod/PointNetVlad-Pytorch

The main differences are:

• data_preprocess folder: Data preprocessing or Tsinghua data and ARS548 data.
• generating_queries folder: Query and database partitioning for Tsinghua data and ARS548 data.
• Some parameters in configs added.
• add gen_libtorch_model.py.
• tools folder: Some scripts for processing data or post-processing results.

Pre-Requisites

• PyTorch 2.0.1

Generate pickle files

cd generating_queries/

# For training tuples
python generate_training_tuples_baseline.py

# For network evaluation
python generate_test_sets.py

Train

python train_pointnetvlad.py

Evaluate

python evaluate_radar.py

Evaluate with pr curve

python evaluate_radar_PR.py

Take a look at train_pointnetvlad.py and evaluate.py for more parameters