Semantic Scene Completion (SSC) constitutes a pivotal element in autonomous driving perception systems, tasked with inferring the 3D semantic occupancy of a scene from sensory data. To improve accuracy, prior research has implemented various computationally demanding and memory- intensive 3D operations, imposing significant computational requirements on the platform during training and testing. This paper proposes L2COcc, a lightweight camera-centric SSC framework that also accommodates LiDAR inputs. With our proposed efficient voxel transformer (EVT) and cross-modal knowledge modules, including feature similarity distillation (FSD), TPV distillation (TPVD) and prediction alignment dis- tillation (PAD), our method substantially reduce computational burden while maintaining high accuracy. The experimental evaluations demonstrate that our proposed method surpasses the current state-of-the-art vision-based SSC methods regarding accuracy on both the SemanticKITTI and SSCBench-KITTI- 360 benchmarks, respectively. Additionally, our method is more lightweight, exhibiting a reduction in both memory consump- tion and inference time by over 23% compared to the current state-of-the-arts method.

More video demonstrations can be found at the project page.

The overall framework of our proposed L2COcc, comprised of three stages: voxel feature generation (indicated by the gray background), TPV-based occupancy prediction network (indicated by the blue background), and cross-modal knowledge distillation (indicated by the green background).

step 1. Refer to install.md to install the environment.

step 2. Refer to dataset.md to prepare SemanticKITTI and KITTI360 dataset.

step 3. Refer to train_and_eval.md for training and evaluation.

step 4. Refer to visualization.md for visualizations.

1. SemanticKITTI

   Model	Sensor	Split	IoU	mIoU	Download
   L2COcc-C	Camera	val	45.56	16.72	ckpt/log
   L2COcc-C	Camera	test	44.31	17.03	output
   L2COcc-D	Camera	val	45.30	18.22	ckpt/log
   L2COcc-D	Camera	test	45.37	18.18	output
   L2COcc-L	LiDAR	val	60.66	24.21	ckpt/log
   L2COcc-L	LiDAR	test	60.32	23.37	output

2. KITTI360

   Model	Sensor	Split	IoU	mIoU	Download
   L2COcc-C	Camera	test	48.07	20.11	ckpt/log
   L2COcc-D	Camera	test	48.83	20.99	ckpt/log
   L2COcc-L	LiDAR	test	57.60	25.22	ckpt/log

Many thanks to these exceptional open source projects:

• CGFormer
• LiCROcc
• PointOcc
• BEVFormer
• mmdet3d
• MonoScene
• semantic-kitti-api
• MobileStereoNet
• Symphonize
• DFA3D
• VoxFormer

If you find our work beneficial for your research, please consider citing our paper and give us a star:

@misc{wang2025l2cocclightweightcameracentricsemantic,
      title={L2COcc: Lightweight Camera-Centric Semantic Scene Completion via Distillation of LiDAR Model}, 
      author={Ruoyu Wang and Yukai Ma and Yi Yao and Sheng Tao and Haoang Li and Zongzhi Zhu and Yong Liu and Xingxing Zuo},
      year={2025},
      eprint={2503.12369},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.12369}, 
}

If you encounter any issues, please contact samuraiwry@gmail.com.