Scribble2Scene

Song Wang, Jiawei Yu, Wentong Li, Hao Shi, Kailun Yang, Junbo Chen*, Jianke Zhu*

This is the official implementation of Label-efficient Semantic Scene Completion with Scribble Annotations (IJCAI 2024) [Paper].

Getting Started

We provide the core codes of our proposed Scribble2Scene for online model training (Stage-II):

./code
    └── projects/
    │       ├── configs/
    │       │     ├── scribble2scene/
    |       |     |          ├──scribble2scene-distill.py  # the config file for Scribble2Scene Stage-II
    │       ├── mmdet3d_plugin/
    │       │     ├── scribble2scene/
    |       |     |          ├──detectors
    |       |     |          |    ├──scribble2scene_distill.py  # our Teacher-Labeler and online model architecture
    |       |     |          ├──dense_heads
    |       |     |          |    ├──scribble2scene_head.py  # our used completion head and loss functions
    |       |     |          ├──utils
    |       |     |          |    ├──distillation_loss.py  # our proposed range-guided offline-to-online distillation loss
    └──tools/

Prepare Data-SemanticKITTI

Direct downloading:

• The semantic scene completion dataset v1.1 (SemanticKITTI voxel data, 700 MB) from SemanticKITTI website.
• The RGB images (Download odometry data set (color, 65 GB)) from KITTI Odometry website.

Run and Eval

Train the online model with our proposed Scribble2Scene on 4 GPUs

./tools/dist_train.sh ./projects/configs/scribble2scene/scribble2scene-distill.py 4

Eval the online model with our proposed Scribble2Scene on 4 GPUs

./tools/dist_test.sh ./projects/configs/scribble2scene/scribble2scene-distill.py ./path/to/ckpts.pth 4

Acknowledgement

Many thanks to these excellent open source projects:

• VoxFormer
• mmdetection3d