Gaussian similarity-based adaptive dynamic label assignment for tiny object detection

Ronghao Fu, Chengcheng Chen, Shuang Yan, Ali Asghar Heidari, Xianchang Wang, José Escorcia-Gutierrez, Romany Fouad Mansour, Huiling Chen

Abstract: Highlights•GPM is presented to measure the similarity of tiny objects.•GPM outperforms other IoU-based metrics.•The ADAS is proposed to solve the inter-sample and intra-sample imbalance.•The sufficient supervision information for network has been guaranteed by ADAS-GPM.•ADAS-GPM achieves state-of-the-art performance on AI-TODv2.