3D Point Cloud Understanding, Contrastive Learning

Environments:

• Python 3.7
• Cuda 10.2
• torch 1.8.1
• torchvision 0.9.1
• timm 0.6.12

Train steps:

• Step 1: run trainer.py to pretrain the model with ShapeNet dataset;
• Step 2: run cls_eval_svm.py to get linear classification results;
• Step 3: run cls_eval_fsl.py to get Few-Shot-Learning results;
• Step 4: run cls_finetune.py to get finetune classification results;
• Step 5: run tSNE_plot to get t-SNE figure;

Dataset:

• shapenet dataset
• modelnet40 dataset
• scanobjectnn dataset
• shapenet_part dataset

Events file:

• pretrain events
• local finetune events

Weights file:

• pretrained model weights
• finetune model weights