kentechx / shapenet_part_experiments

Experiments on ShapeNetPart dataset

shapenet_part_experiments

This repository contains part segmentation experiments conducted on the ShapeNetPart dataset.

Dataset

The ShapeNetPart dataset is annotated for 3D object part segmentation. It consists of 16,880 models from 16 shape categories, with 14,006 3D models for training and 2,874 for testing. The number of parts for each category is between 2 and 6, with 50 different parts in total.

Install Dependencies

pip install -r requirements.txt

Experiments

The experiments are conducted on the following models:

• DGCNN
• PointNet
• PointNet2

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 processed dataset will be downloaded automatically when running the training scripts.

Results

The table below presents the instance mIoU and class mIoU of the models on the ShapeNetPart dataset with 2048 points.

Model	input	ins. mIoU	cls. mIoU	device
PointNet2SSG	xyz	84.8%	82.0%	1x 3090
PointNet2MSG	xyz	85.2%	82.5%	1x 3090
DGCNN	xyz	85.4%	83.1%	1x 3090

You can reproduce the results by running the corresponding scripts in the code folder with default configurations. For example, to train the PointNet2 model, run the following command

python code/train_pointnet2.py