Home

Awesome

DKNet (Dynamic Kernel Network)

3D Instances as 1D Kernels (ECCV2022)

overview

Code for the paper 3D Instances as 1D Kernels, ECCV 2022.

Authors: Wu Yizheng, Shi Min, Du Shuaiyuan, Lu Hao, Cao Zhiguo, Zhong Weicai

[paper]

Introduction

We introduce a 3D instance representation, termed instance kernels, where instances are represented by one-dimensional vectors that encode the semantic, positional, and shape information of 3D instances. We show that instance kernels enable easy mask inference by simply scanning kernels over the entire scenes, avoiding the heavy reliance on proposals or heuristic clustering algorithms in standard 3D instance segmentation pipelines. The idea of instance kernel is inspired by recent success of dynamic convolutions in 2D/3D instance segmentation. However, we find it non-trivial to represent 3D instances due to the disordered and unstructured nature of point cloud data, e.g., poor instance localization can significantly degrade instance representation. To remedy this, we construct a novel 3D instance encoding paradigm. First, potential instance centroids are localized as candidates. Then, a candidate merging scheme is devised to simultaneously aggregate duplicated candidates and collect context around the merged centroids to form the instance kernels. Once instance kernels are available, instance masks can be reconstructed via dynamic convolutions whose weights are conditioned on instance kernels. The whole pipeline is instantiated with a dynamic kernel network (DKNet). Results show that DKNet outperforms the state of the arts on both ScanNetV2 and S3DIS datasets with better instance localization.

Installation

Requirements

Virtual Environment

conda create -n dknet python==3.7
source activate dknet

Install DKNet

(1) Clone the DKNet repository.

git clone https://github.com/W1zheng/DKNet.git --recursive
cd DKNet

(2) Install the dependent libraries.

pip install -r requirements.txt
conda install -c bioconda google-sparsehash

(3) Install spconv

We use spconv2.x. Please refer to spconv for details.

(4) Compile the external C++ and CUDA ops.

cd ./lib/dknet_ops
export CPLUS_INCLUDE_PATH={conda_env_path}/dknet/include:$CPLUS_INCLUDE_PATH
python setup.py build_ext develop

{conda_env_path} is the location of the created conda environment, e.g., /anaconda3/envs. Alternative installation guide can be found in here.

Build example:

cd ./lib/segmentator

cd csrc && mkdir build && cd build

cmake .. \
-DCMAKE_PREFIX_PATH=`python -c 'import torch;print(torch.utils.cmake_prefix_path)'` \
-DPYTHON_INCLUDE_DIR=$(python -c "from distutils.sysconfig import get_python_inc; print(get_python_inc())")  \
-DPYTHON_LIBRARY=$(python -c "import distutils.sysconfig as sysconfig; print(sysconfig.get_config_var('LIBDIR'))") \
-DCMAKE_INSTALL_PREFIX=`python -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())'`

make && make install # after install, please do not delete this folder (as we only create a symbolic link)

Further information can be found in here.

Data Preparation

(1) Download the ScanNet v2 dataset.

(2) Put the data in the corresponding folders.

The dataset files are organized as follows.

DKNet
├── dataset
│   ├── scannetv2
│   │   ├── train
│   │   │   ├── [scene_id]_vh_clean_2.ply & [scene_id]_vh_clean_2.labels.ply & [scene_id]_vh_clean_2.0.010000.segs.json & [scene_id].aggregation.json
│   │   ├── val
│   │   │   ├── [scene_id]_vh_clean_2.ply & [scene_id]_vh_clean_2.labels.ply & [scene_id]_vh_clean_2.0.010000.segs.json & [scene_id].aggregation.json
│   │   ├── test
│   │   │   ├── [scene_id]_vh_clean_2.ply
│   │   ├── scannetv2-labels.combined.tsv

(3) Generate input files [scene_id]_inst_nostuff.pth for instance segmentation.

cd dataset/scannetv2
python prepare_data_inst.py --data_split train
python prepare_data_inst.py --data_split val
python prepare_data_inst.py --data_split test

Training

CUDA_VISIBLE_DEVICES=0 python train.py --config config/DKNet_run1_scannet.yaml

You can start a tensorboard session by

tensorboard --logdir=./exp --port=6666

Inference and Evaluation

(1) If you want to evaluate on validation set, prepare the .txt instance ground-truth files as the following.

cd dataset/scannetv2
python prepare_data_inst_gttxt.py

Make sure that you have prepared the [scene_id]_inst_nostuff.pth files before.

(2) Test and evaluate.

a. To evaluate on validation set, set split and eval in the config file as val and True. Then run

CUDA_VISIBLE_DEVICES=0 python test.py --config config/DKNet_run1_scannet.yaml

An alternative evaluation method is to set save_instance as True, and evaluate with the ScanNet official evaluation script.

b. To run on test set, set (split, eval, save_instance) as (test, False, True). Then rund

CUDA_VISIBLE_DEVICES=0 python test.py --config config/DKNet_run1_scannet.yaml

c. To test with a pretrained model, run

CUDA_VISIBLE_DEVICES=0 python test.py --config config/DKNet_run1_scannet.yaml --pretrain $PATH_TO_PRETRAIN_MODEL$

We provide a pretrained model trained on ScanNet v2 dataset. Download it here. Its performance on ScanNet v2 validation set is 50.4/66.9/75.9 in terms of mAP/mAP50/mAP25.

Visualize

To visualize the point cloud, you should first install mayavi. Then you could visualize by running

cd util
python visualize.py --data_root $DATA_ROOT$ --result_root $RESULT_ROOT$ --room_name $ROOM_NAME$ --room_split $ROOM_SPLIT$ --task $TASK$

The visualization task could be input, instance_gt, instance_pred, semantic_pred, semantic_gt, candidate and candidate_merge.

Results on ScanNet Benchmark

Quantitative results on ScanNet test set at the submisison time.

scannet_result

TODO List

Citation

If you find this work useful in your research, please cite:

@inproceedings{wu20223d,
  title={3D Instances as 1D Kernels},
  author={Wu, Yizheng and Shi, Min and Du, Shuaiyuan and Lu, Hao and Cao, Zhiguo and Zhong, Weicai},
  booktitle={Computer Vision--ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23--27, 2022, Proceedings, Part XXIX},
  pages={235--252},
  year={2022},
  organization={Springer}
}

Acknowledgement

This repo is built upon several repos, e.g., spconv, PointGroup and DyCo3D.

Contact

If you have any questions or suggestions about this repo, please feel free to contact me (yzwu21@hust.edu.cn).