Gaozhongpai/PaiConvPoint

Official repository for the paper "PAI-Conv: Permutable Anisotropic Convolutional Networks for Learning on Point Clouds" [Classification for uniformly sampled data]

PAI-Conv: Permutable Anisotropic Convolutional Networks for Learning on Point Clouds [Classification]

Point Cloud Classification

• Run the training script:

python main.py --exp_name=paigcnn_1024 --model=paigcnn --num_points=1024 --k=20 --use_sgd=True

python main.py --exp_name=paigcnn_2048 --model=paigcnn --num_points=2048 --k=40 --use_sgd=True

• Run the evaluation script after training finished:

python main.py --exp_name=paigcnn_1024_eval --model=paigcnn --num_points=1024 --k=20 --use_sgd=True --eval=True --model_path=checkpoints/paigcnn_1024/models/model.t7

python main.py --exp_name=paigcnn_2048_eval --model=paigcnn --num_points=2048 --k=40 --use_sgd=True --eval=True --model_path=checkpoints/paigcnn_2048/models/model.t7

• Run the evaluation script with pretrained models:

python main.py --exp_name=paigcnn_1024_eval --model=paigcnn --num_points=1024 --k=20 --use_sgd=True --eval=True --model_path=pretrained/model.1024.t7

python main.py --exp_name=paigcnn_2048_eval --model=paigcnn --num_points=2048 --k=40 --use_sgd=True --eval=True --model_path=pretrained/model.2048.t7

Data Organization

The following is the organization of the dataset for 8192 points expected by the code:

* dataset/
    * train/ (created by data_generation.py)
    * test/ (created by data_generation.py)

Acknowlegements:

The structure of this codebase is borrowed from DGCNN.