상세정보
학술대회
Intra-Batch Features Separation for Indoor and Outdoor Pedestrian-View Intersection Classification
- 발행일
- 202110
- 출처
- International Conference on Control, Automation and Systems (ICCAS) 2021, pp.1-4
- 협약과제
- 21HS1600, 불확실한 지도 기반 실내ㆍ외 환경에서 최종 목적지까지 이동로봇을 가이드할 수 있는 AI 기술 개발, 이재영
- 초록
- Pedestrian-view intersection classification is an important component to assist robots to navigate in the pedestrian path. To solve this problem, previous approaches simply fine-tune an ImageNet-pretrained network with intersection classification dataset using cross-entropy loss as classification loss in an end-to-end manner. In this work, we propose a novel additional loss to further improve the model's capability to discriminate intersection and non-intersection class. This loss is directly calculated on the features in a given mini-batch without requiring any additional inference. Furthermore, previous works cover only outdoor domain while we also propose indoor domain in addition to the outdoor intersection classification dataset. Extensive experiments show that the models trained using the proposed loss yields better performance compared to the models trained without it on both indoor and outdoor datasets. This demonstrates the potential of the proposed loss in improving the discrimination capability of our models.
- KSP 제안 키워드
- Additional loss, Cross-Entropy, End to End(E2E), Entropy loss