Jittor_warm-up_contest

A Jittor implementation of Point Cloud Transformer (PCT) for ModelNet40 classification.

Project Description

This project implements Point Cloud Transformer (PCT) using the Jittor deep learning framework for 3D point cloud classification on the ModelNet40 dataset.

Requirements

• Python 3.10
• Jittor 1.3.11.0
• NumPy

Installation

conda create -n pa3 python=3.10 -y
conda activate pa3
pip install jittor

Dataset

The ModelNet40 dataset files are provided by the Educoder competition platform:https://www.educoder.net/competitions/Jittor-7.

Please place the following files in the data/ directory:

• train_points.npy
• train_labels.npy
• test_points.npy

Directory structure:

data/
├── train_points.npy
├── train_labels.npy
└── test_points.npy

Usage

python pct.py --epochs 200 --batch_size 32 --lr 0.001