Density-Guided Dense Pseudo Label Selection For Semi-supervised Oriented Object Detection
This work addresses a domain-specific challenge in semi-supervised oriented object detection for aerial imagery, offering an incremental improvement over existing methods.
The paper tackles the problem of inefficient pseudo-label selection for multi-oriented and dense objects in aerial scenes by proposing a density-guided method, achieving 49.78 mAP on the DOTA-v1.5 benchmark with only 5% labeled data, surpassing previous state-of-the-art methods.
Recently, dense pseudo-label, which directly selects pseudo labels from the original output of the teacher model without any complicated post-processing steps, has received considerable attention in semi-supervised object detection (SSOD). However, for the multi-oriented and dense objects that are common in aerial scenes, existing dense pseudo-label selection methods are inefficient because they ignore the significant density difference. Therefore, we propose Density-Guided Dense Pseudo Label Selection (DDPLS) for semi-supervised oriented object detection. In DDPLS, we design a simple but effective adaptive mechanism to guide the selection of dense pseudo labels. Specifically, we propose the Pseudo Density Score (PDS) to estimate the density of potential objects and use this score to select reliable dense pseudo labels. On the DOTA-v1.5 benchmark, the proposed method outperforms previous methods especially when labeled data are scarce. For example, it achieves 49.78 mAP given only 5\% of annotated data, which surpasses previous state-of-the-art method given 10\% of annotated data by 1.15 mAP. Our codes is available at https://github.com/Haru-zt/DDPLS.