Efficient Weakly-Supervised Object Detection With Pseudo Annotations

Abstract

Weakly-supervised object detection (WSOD) has attracted lots of attention in recent years. However, there is still a big gap between WSOD and generic object detection. The main barriers to the efficiency of WSOD are the ineffective data augmentations and inaccurate bounding box predictions. Given only the image-level annotations, it's hard for WSOD to effectively utilize variant data augmentations and accurately regress the bounding boxes. Although a fully-supervised object detector can be trained using annotations generated from the weakly-supervised obejct detector, the performance is still severely limited due to the low quality of mined pseudo annotations. This paper proposes an efficient WSOD method with pseudo annotations (EWPA) to make better use of imperfect annotations. With the assistance of pseudo annotations, EWPA can effectively regress more accurate bounding boxes while the traditional WSOD can only locate the salient parts of an object. Furthermore, pseudo annotations can help design more complex data augmentations to drive the network learning more discriminative feature representations. Extensive experiments are conducted on PASCAL VOC 2007 and 2012 datasets and validate the effectiveness of EWPA.

Proposals Generation for Weakly Supervised Object Detection in Artwork Images

“…Interesting approaches are presented by [ 21 , 23 ] and [ 22 ], which optimize the problem by dividing it into sub-parts, respectively, at the data level and loss level. Recent works combine MIL and Deep Neural Networks (DNN) in WSDDNs [ 9 , 24 , 25 , 26 , 27 ]. A typical WSDDN is composed of two streams, devoted to classification and localization trained jointly to mine positive samples [ 28 ].…”

Object detection for panoramic images based on MS‐RPN structure in traffic road scenes

“…Besides, in order to make full use of existing image classification datasets and object detection datasets, and reduce the annotation work amount of object‐level labels and the difficulty of annotating labels, some weakly supervised and semi‐supervised object detection methods are proposed. Such as Ref [24] and Ref [25], in the training stage, only image‐level labels are accessible in the weakly supervised setting, while image‐level labels combined with some object‐level labels are available in the semi‐supervised setting. During the test stage, bounding boxes for new images will be predicted.…”