Proposal-Refined Weakly Supervised Object Detection In Underwater Images

Recently, Convolutional Neural Networks (CNNs) have achieved great success in object detection due to their outstanding abilities of learning powerful features on large-scale training datasets. One of the critical factors of their success is the accurate and complete annotation of the training dataset. However, accurately annotating the training dataset is difficult and time-consuming in some applications such as object detection in underwater images due to severe foreground clustering and occlusion. In this paper, we study the problem of object detection in underwater images with incomplete annotation. To solve this problem, we propose a proposal-refined weakly supervised object detection method, which consists of two stages. The first stage is a weakly-fitted segmentation network for foreground-background segmentation. The second stage is a proposal-refined detection network, which uses the segmentation results of the first stage to refine the proposals and therefore can improve the performance of object detection. Experiments are conducted on the Underwater Robot Picking Contest 2017 dataset (URPC2017) which has 19967 underwater images containing three kinds of objects: sea cucumber, sea urchin and scallop. The annotation of the training set is incomplete. Experimental results show that the proposed method greatly improves the detection performance compared to several baseline methods.