Skip to content

[CVPR 2024] Official PyTorch Code of SeaBird: Segmentation in Bird's View with Dice Loss Improves Monocular 3D Detection of Large Objects

License

Notifications You must be signed in to change notification settings

abhi1kumar/SeaBird

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

22 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SeaBird: Segmentation in Bird's View with Dice Loss Improves Monocular 3D Detection of Large Objects

KITTI-360 Demo | [nuScenes Demo] | Project | Talk | Slides | Poster

PWC PWC

arXiv License: MIT Visitors GitHub Stars

Abhinav Kumar1 · Yuliang Guo2 · Xinyu Huang2 · Liu Ren2 · Xiaoming Liu1
1Michigan State University, 2Bosch Research North America, Bosch Center for AI

in CVPR 2024

Monocular 3D detectors achieve remarkable performance on cars and smaller objects. However, their performance drops on larger objects, leading to fatal accidents. Some attribute the failures to training data scarcity or the receptive field requirements of large objects. In this paper, we highlight this understudied problem of generalization to large objects. We find that modern frontal detectors struggle to generalize to large objects even on nearly balanced datasets. We argue that the cause of failure is the sensitivity of depth regression losses to noise of larger objects. To bridge this gap, we comprehensively investigate regression and dice losses, examining their robustness under varying error levels and object sizes. We mathematically prove that the dice loss leads to superior noise-robustness and model convergence for large objects compared to regression losses for a simplified case. Leveraging our theoretical insights, we propose SeaBird (Segmentation in Bird's View) as the first step towards generalizing to large objects. SeaBird effectively integrates BEV segmentation on foreground objects for 3D detection, with the segmentation head trained with the dice loss. SeaBird achieves SoTA results on the KITTI-360 leaderboard and improves existing detectors on the nuScenes leaderboard, particularly for large objects.

Citation

If you find our work useful in your research, please consider starring the repo and citing:

@inproceedings{kumar2024seabird,
   title={{SeaBird: Segmentation in Bird's View with Dice Loss Improves Monocular $3$D Detection of Large Objects}},
   author={Kumar, Abhinav and Guo, Yuliang and Huang, Xinyu and Ren, Liu and Liu, Xiaoming},
   booktitle={CVPR},
   year={2024}
}

Single Camera (KITTI-360) Models

See PanopticBEV

Model Zoo

We provide logs/models/predictions for the main experiments on KITTI-360 Val /KITTI-360 Test data splits available to download here.

Data_Splits Method Config
(Run)
Weight
/Pred
Metrics Lrg
(50)
Car
(50)
Mean
(50)
Lrg
(25)
Car
(25)
Mean
(25)
Lrg
Seg
Car
Seg
Mean
Seg
KITTI-360 Val Stage 1 seabird_val_stage1 gdrive IoU - - - - - - 23.83 48.54 36.18
KITTI-360 Val PBEV+SeaBird seabird_val gdrive AP 13.22 42.46 27.84 37.15 52.53 44.84 24.30 48.04 36.17
KITTI-360 Test PBEV+SeaBird seabird_test gdrive AP - - 4.64 - - 37.12 - - -

Multi-Camera (nuScenes) Models

See HoP

Model Zoo

nuScenes Val Results

Model Resolution Backbone Pretrain APLrg mAP NDS Ckpt/Log/Pred
HoP_BEVDet4D_256 256x704 ResNet50 ImageNet-1K 0.274 0.399 0.509 ckpt / log
HoP+SeaBird_256 Stage1 256x704 ResNet50 ImageNet-1K - - - gdrive
HoP+SeaBird_256 256x704 ResNet50 ImageNet-1K 0.282 0.411 0.515 gdrive
HoP+SeaBird_512 Stage1 512x1408 ResNet101 ImageNet-1K - - - gdrive
HoP+SeaBird_512 512x1408 ResNet101 ImageNet-1K 0.329 0.462 0.547 gdrive
HoP+SeaBird_640 Stage1 640x1600 V2-99 DDAD15M - - - gdrive
HoP+SeaBird_640 640x1600 V2-99 DDAD15M 0.403 0.527 0.602 gdrive

nuScenes Test Results

Model Resolution Backbone Pretrain APLrg mAP NDS Ckpt/Log/Pred
HoP+SeaBird_512 Test 512x1408 ResNet101 ImageNet-1K 0.366 0.486 0.570 gdrive
HoP+SeaBird_640 Val 640x1600 V2-99 DDAD15M 0.384 0.511 0.597 gdrive

Acknowledgements

We thank the authors of the following awesome codebases:

Please also consider citing them.

Contributions

We welcome contributions to the SeaBird repo. Feel free to raise a pull request.

↳ Stargazers

Stargazers repo roster for @nastyox/Repo-Roster

↳ Forkers

Forkers repo roster for @nastyox/Repo-Roster

License

SeaBird code is under the MIT license.

Contact

For questions, feel free to post here or drop an email to this address- [email protected]