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].
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/
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.
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
Many thanks to these excellent open source projects: