Self-Guided Proposal Generation for Weakly Supervised Object Detection
Gong Cheng,Xuan Xie,Weining Chen,Xiaoxu Feng,Xiwen Yao,Junwei Han
DOI: https://doi.org/10.1109/TGRS.2022.3181466
IF: 8.2
2022-01-01
IEEE Transactions on Geoscience and Remote Sensing
Abstract:Weakly supervised object detection (WSOD) in remote sensing images remains a challenging task when learning object detectors with only image-level labels. As we know, object proposal generation plays a crucial role in WSOD. At present, the proposal generation of most existing WSOD methods mainly relies on heuristic strategies such as selective search and Edge Boxes. However, the proposals obtained by the above methods cannot well cover the entire objects, severely hindering the performance of WSOD. To address this issue, this article proposes a Self-guided Proposal Generation approach, termed SPG. It can be easily implemented with most WSOD methods in a unified framework. To this end, we first introduce a confidence propagation approach to obtain the objectness confidence map for each image, which, on the one hand, highlights informative object locations and, on the other hand, aggregates discriminative feature representation by combining the objectness confidence map with the deep features. Then, the proposal generation is implemented by mining informative regions as proposals on the objectness confidence map. Extensive evaluations on two challenging datasets demonstrate that our SPG significantly improves the baseline methods, online instance classifier refinement (OICR) and min-entropy latent model (MELM), by large margins (for OICR: 15.86% mAP and 12.89% CorLoc gains on the NWPU VHR-10.v2 dataset and 3.65% mAP and 4.87% CorLoc gains on the DIOR dataset; for MELM: 20.51% mAP and 23.54% CorLoc gains on the NWPU VHR-10.v2 dataset and 7.11% mAP and 4.96% CorLoc gains on the DIOR dataset) and achieves the state-of-the-art results compared with existing methods.