Skip to content

Latest commit

 

History

History
48 lines (36 loc) · 2.09 KB

README.md

File metadata and controls

48 lines (36 loc) · 2.09 KB

Semantic Segmentation

This repo contains the code and configuration files for point cloud semantic segmentation.

Related Papers

  • A Unified Query-based Paradigm for Point Cloud Understanding (paper)

Results and Models

Note. All models below are trained with 8 1080TI GPU, follow EQ-Paradigm, and use a Q-Net to enable a free combination between backbones and heads. (*) means the improvement compared to the model with its original backbone network without Q-Net.

ScanNet Semantic Segmentation Model

Backbone mIoU mAcc allAcc download
EQNet SparseConvNet 75.1 (+2.2) 82.7 (+1.9) 91.1 (+0.7) model

S3DIS Semantic Segmentation Model

To be released soon.

More models

Performance of other backbones supported in this codebase will be released soon.

Getting Started

ScanNet V2

  • Data preparation: Download ScanNet v2 here and preprocess the data.

    cd /path/to/DeepVision3D/DVSegmentation/data/scannetv2
python prepare_data.py --scannet_path /path/to/ScanNet --split [train/val/test]

  • Training: You can train on ScanNet v2 with following codes.

    cd /path/to/DeepVision3D/DVSegmentation
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 bash train_segmentation.sh 8 --config config/eqnet_scannet.yaml

  • Testing: 

    python test_segmentation.py --config config/eqnet_scannet.yaml --set NECK.QUERY_POSITION_CFG.SELECTION_FUNCTION _get_point_query_position

    For testing our provided model:

    CHECKPOINT=/path/to/eqnet_scannet_v2-000000600.pth
python test_segmentation.py --config config/eqnet_scannet.yaml --pretrain ${CHECKPOINT} --set NECK.QUERY_POSITION_CFG.SELECTION_FUNCTION _get_point_query_position