BROWSE
Detail
Conference Paper
다중 공개 데이터셋을 이용한 자기지도 단안 깊이 학습
Cited - time in 
Share 
Share
- Issue Date
- 2023-11
- Citation
- 한국자동차공학회 학술 대회 (추계) 2023, pp.1513-1516
- Publisher
- 한국자동차공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- Self-supervised learning enables the depth estimation network to be trained by simple driving data without the need for ground-truth depth. In this study, we present a method to simultaneously use five types of public datasets (SeasonDepth, KITTI, nuScenes, Waymo Open, DDAD) for learning. Since each dataset was collected in a different environment, the learned depth estimation network is more robust to environmental change. Our model shows competitive results on the SeasonDepth benchmark. This work shows that it is possible to improve the performance of existing algorithms by simply using more datasets.
- KSP Keywords
- Depth estimation, Environmental change, Public Datasets, driving data, ground truth, need for, self-supervised learning