Exploring Weakly Labeled Images for Video Object Segmentation With Submodular Proposal Selection

Abstract

Video object segmentation (VOS) is important for various computer vision problems, and handling it with minimal human supervision is highly desired for the large-scale applications. To bring down the supervision, existing approaches largely follow a data mining perspective by assuming the availability of multiple videos sharing the same object categories. It, however, would be problematic for the tasks that consume a single video. To address this problem, this paper proposes a novel approach that explores weakly labeled images to solve video object segmentation. Given a video labeled with a target category, images labeled with the same category are collected, from which noisy object exemplars are automatically discovered. After that the proposed approach extracts a set of region proposals on various frames and efficiently matches them with massive noisy exemplars in terms of appearance and spatial context. We then jointly select the best proposals across the video by solving a novel submodular problem that combines region voting and global region matching. Finally, the localization results are leveraged as strong supervision to guide pixel-level segmentation. Extensive experiments are conducted on two challenging public databases: Youtube-Objects and DAVIS. The results suggest that the proposed approach improves over previous weakly supervised/unsupervised approaches significantly, showing a performance even comparable with the several approaches supervised by the costly manual segmentations.

CATRO: Channel Pruning via Class-Aware Trace Ratio Optimization

“…In recent years, the combinatorial selection based on submodular functions has been one of the most promising methods in machine learning and data mining. It has been a surge of interest in lots of computer tasks, such as visual recognition [31], segmentation [32], clustering [33], [34], active learning [35] and user recommendation [36]. The submodular set function maximization [37], which is to maximize a submodular function, is one of the most basic and important problem.…”

Object Detection Based on Faster R-CNN Algorithm with Skip Pooling and Fusion of Contextual Information

“…Cheng et al [ 27 ] proposed a proposal generation method, which can generate more proposals that have higher intersection over union (IoU) with ground truth boxes than those obtained by greedy search approaches, which can better envelop entire objects. Zhang et al [ 28 ] proposed a fast matching algorithm that robustly matches region proposals with massive exemplars in terms of appearance and spatial context, and can robustly handle noisy localizations of image exemplars. Compared with two-stage algorithms in terms of detection speed, one-stage algorithms are faster, but in terms of detection accuracy, the popular version of the former is better than the latter.…”

Team Composition in PES2018 Using Submodular Function Optimization

“…In recent years, the submodular function optimization has been seen in many machine learning and computer vision applications and is usually applied to coverage issues such as video segmentation [20], [21], document summarization [22], advertisement allocation [23], and information gathering [24]. In this article, we will convert the team recommendation problem into a submodular function optimization with the constraint of salary cost, and seek for greedy algorithm based solutions to the problem.…”