Object-Aware Instance Labeling for Weakly Supervised Object Detection

Abstract

Weakly supervised object detection (WSOD), where a detector is trained with only image-level annotations, is attracting more and more attention. As a method to obtain a well-performing detector, the detector and the instance labels are updated iteratively. In this study, for more efficient iterative updating, we focus on the instance labeling problem, a problem of which label should be annotated to each region based on the last localization result. Instead of simply labeling the top-scoring region and its highly overlapping regions as positive and others as negative, we propose more effective instance labeling methods as follows. First, to solve the problem that regions covering only some parts of the object tend to be labeled as positive, we find regions covering the whole object focusing on the context classification loss. Second, considering the situation where the other objects contained in the image can be labeled as negative, we impose a spatial restriction on regions labeled as negative. Using these instance labeling methods, we train the detector on the PASCAL VOC 2007 and 2012 and obtain significantly improved results compared with other state-ofthe-art approaches.

Deep Learning for Weakly-Supervised Object Detection and Object Localization: A Survey

“…It selects positive proposals by leveraging the loss of context regions. For example, OAILWSD [50] believes that a proposal not tightly covers the object instance if the loss of the context feature maps of this proposal tends to decrease. Thus, OAILWSD first leverages the context classification loss to label regions.…”

Deep neural networks based visual object detection in videos

“…To solve the non-convexity problem in weakly supervised loss function, C-MIL [102] proposes an optimization algorithm in a continuation way. The approach proposed in [103] proposes a method to label objects in instance-level using spatial constrain. Most existing weakly supervised object detection methods are proposed for images.…”

“…Our OPG is also different from methods [1,103]. Although methods [1,103] assign 0 weight for samples with low IOU value, they do not limit the number of training proposals. They simply set a fixed threshold value to filter out some samples.…”

“…Some people may mix up our proposed method with other weakly supervised object detection methods [1,100,103,116] Although [116] and [100] try to filter out background samples, they need the segmentation results to filter out these samples.…”

“…training process in [1,103] still involves many negative samples. By contrast, our methods limit the number of proposals according to different training phases (DPC).…”