This repository contains experiments conducted on the ModelNet40 dataset. The ModelNet40 dataset is a widely used benchmark dataset for 3D classification tasks.
The ModelNet40 dataset consists of 12,311 CAD models from 40 object categories, with approximately 300 instances per
category. The dataset is divided into two sets: a training set of 9,843 models and a test set of 2,468 models. The
original CAD models are provided in the .off
format.
The experiments are conducted on the processed data in .hdf5
format. Each processed object contains 2048 points
sampled from the original CAD models.
pip install -r requirements.txt
The experiments are conducted on the following models:
Training models by running the corresponding scripts in the code
folder. For example, to train the DGCNN model, run
the following command:
python code/train_dgcnn.py
The table below presents the classification accuracy of the models on the ModelNet40 dataset (1024 points, trained on a single Nvidia RTX 3090 GPU). Some of the following results are the best outcomes obtained after hyperparameter sweepings.
Model | input | Overall Accuracy | sweep |
---|---|---|---|
DGCNN | xyz | 91.7% | ✓ |
PointNet | xyz | 90.7% | ✓ |
PointNet2SSG | xyz | 90.7% | |
PointNet2MSG | xyz | 92.1% | |
PointNext | xyz | 91.3% |
You can reproduce the results by running the corresponding scripts in the code
folder with default configurations.
For example, to train the PointNet model, run the following command
python code/train_pointnet.py