Weakly supervised foreground learning for weakly supervised localization and detection. (May 2023)

• Record Type: Journal Article
• Authors: Zhang, Chen-Lin; Li, Yin; Wu, Jianxin
• Is Part Of: Pattern recognition. Volume 137(2023)
• Publication Date: 2023-05
• Subjects: Weakly supervised object localization -- Weakly supervised object detection -- Foreground learning; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.109279
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We find that groundtruth foreground masks can greatly benefit tasks such as WSOL and WSOD. A previous literature claimed that WSOL is an ill-posed problem with classification models and classification labels. We propose WSFL (weakly supervised foreground learning) which learns foreground masks in a weakly supervised fashion (i.e., no extra supervision). With the help of the WSFL task, we can avoid the ill-posed problem in WSOL. Applying WSFL to WSOL and WSOD, our method establishes new state-of-the-art results for both tasks. We achieve 74.37% Top-1 localization accuracy on CUB-200 and we achieve 55.7% mAP on VOC2007. Abstract: Modern deep learning models require large amounts of accurately annotated data, which is often difficult to satisfy. Hence, weakly supervised tasks, including weakly supervised object localization (WSOL) and detection (WSOD), have recently received attention in the computer vision community. In this paper, we motivate and propose the weakly supervised foreground learning (WSFL) task by showing that both WSOL and WSOD can be greatly improved if groundtruth foreground masks are available. More importantly, we propose a complete WSFL pipeline with low computational cost, which generates pseudo boxes, learns foreground masks, and does not need any localization annotations. With the help of foreground masks predicted by our WSFL model, we achieve 74.37% correct localization accuracy on CUB for WSOL, and 55.7% mean average precision on VOC07 for WSOD, thereby establish new state-of-the-art for both tasks. Our WSFL model also shows excellent transfer ability.