Authors: Tianchen Zhao, Xuefei Ning, Ke Hong, Zhongyuan Qiu, Pu Lu, Yali Zhao, Linfeng Zhang, Lipu Zhou, Guohao Dai, Huazhong Yang, Yu Wang
This is the implemention of the ICCV23 paper: Ada3D, For more information, please refer to our Project Page at https://a-suozhang.xyz/ada3d.github.io/
The code is tested on the environment listed below:
torch=1.9.1+cu111
spconv=1.2
openpcdet=0.6.0+633bd6b
CUDA=11.1
RTX3090 GPU.
(1) Setup the Environment :
pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
follow the installation guide to install SPConv-V1.2
local install sparse_bev_tools
cd SparseBEVTools
python setup.py develop
pip install -r requirements.txt
python setup.py develop
(2) Prepare the Data:
ln -s $KITTI_PATH ./data/kitti
(3) Download the Pre-trained Model:
./tools/pretrained-models/
(More pre-trained models on the way)
enter the ./tools
folder
train the model with Sparsity-Preserving BN:
python train.py --cfg_file ./cfgs/masked_bn.yaml --extra_tag masked_bn
(the distributed data parallel training is supported, with the same input args of the command example:)
CUDA_VISIBLE_DEVICES=0,1 bash scripts/dist_train.sh 2 --cfg_file ./cfgs/masked_bn.yaml --extra_tag masked_bn
python train.py --cfg_file ./cfgs/predictor_masked_bn.yaml --extra_tag predictor_train --ckpt ../output/cfgs/masked_bn/default/ckpt/latest_model.pth
python train.py --cfg_file ./cfgs/predictor_masked_bn_tune.yaml --extra_tag predictor_tune --ckpt ../output/cfgs/predictor/default/ckpt/latest_model.pth --hard_drop
python test.py --cfg ./cfgs/predictor_masked_bn_tune.yaml --ckpt ./pretrained-models/checkpoint.pth --hard_drop
This project relies on code and libraries from other open-source projects. We want to express our gratitude to the following developers and projects:
If you find our work is useful in your research, please consider citing:
@article{Zhao2023Ada3DE,
title={Ada3D : Exploiting the Spatial Redundancy with Adaptive Inference for Efficient 3D Object Detection},
author={Tianchen Zhao and Xuefei Ning and Ke Hong and Zhongyuan Qiu and Pu Lu and Yali Zhao and Linfeng Zhang and Lipu Zhou and Guohao Dai and Huazhong Yang and Yu Wang},
journal={ArXiv},
year={2023},
volume={abs/2307.08209},
url={https://api.semanticscholar.org/CorpusID:259937318}
}