RIF

This is the official implementation of the paper "A Lightweight 3D Anomaly Detection Method with Rotationally Invariant". This work has been submitted for review to Elsevier.

The original paper will be released after the DOI number is generated.

Datasets and Pretrain

We use Real3d-ad and Anomaly-ShapeNet for evluation.

Navigate to the training directory and run the training scripts for pre-training:

cd train
./train  # Execute training scripts

The training files contain the complete pre-training pipeline

Run

You need run

python3 main.py --gpu 0 --seed 42 --memory_size 10000 --anomaly_scorer_num_nn 1 --faiss_on_gpu --faiss_num_workers 8 sampler -p 0.1 approx_greedy_coreset