ATSS Paper [https://arxiv.org/abs/1912.02424]
[Abstract]
- Point out that the essential difference between anchor-based and anchor-free detection is actually how to define positive and negative training samples, which leads to the performance gap between them
- Propose an Adaptive Training Sample Selection (ATSS) to automatically select positive and negative samples according to statistical characteristics of object.
- Achieving state-of-the-art performance on MS COCO(AP 50.7%) without introducing any additional overhead.
ATSS 알고리즘이 어려운 개념이 들어가진 않아 이해하기 쉬웠고 성능도 준수하여 흥미로웠던 논문.
