Weakly Supervised Instance Segmentation Based on Two-Stage Transfer Learning

Abstract

Weakly supervised instance segmentation, which could greatly decrease financial and time cost, is one of fundamental computer vision tasks. State-of-the-art methods mainly concentrate on improving the quality of generated pixel level labels, namely masks, using complex traditional segmentation methods, and ignore the effect of the quality of generated masks. Namely, the masks of small object instances tend to be invalid, which would degrades the performance of instance segmentation. In this paper, we propose a twostage transfer learning framework for weakly supervised instance segmentation. We explicitly discriminate the invalid and valid generated masks, and just utilize the valid masks for training to avoid the interference of invalid ones. We use a network-based transfer learning strategy to effectively utilize all useful information, including category labels and bounding-box information of all objects and valid generated masks. Besides, we further use a feature-mapping-based transfer learning strategy to improve the performance of small object instance segmentation. We demonstrate the effectiveness of the proposed method on the PASCAL VOC 2012, and the experimental results show that our proposed method is effective and outperforms state-of-the-art methods.

BLADE: Box-Level Supervised Amodal Segmentation through Directed Expansion

“…WSIS (Arun, Jawahar, and Kumar 2020) builds an annotation consistency framework. In (Sun et al 2020), a two-stage transfer learning framework employing valid generated masks from GrabCut (Rother, Kolmogorov, and Blake 2004) is designed. Recently, BoxInst (Tian et al 2021) raises an observation that proximal pixels with similar colors tend to have the same label, and proposes a concise and effective method incorporating a projection loss and a pairwise loss.…”

Real-Time Instance Segmentation of Traffic Videos for Embedded Devices

“…The work from [ 20 ] was motivated by plant image analysis in the context of plant phenotyping and proposed an exemplar-based recursive instance segmentation framework. In [ 21 ], the authors proposed a two-stage transfer learning framework for weakly supervised instance segmentation, where the algorithms explicitly discriminate between invalidly and validly generated masks and, in training, only make use of the valid masks to avoid the interference of invalid ones. In [ 22 ], the authors studied the problem of aggregating the image-level information of all training images into a large knowledge graph and exploiting semantic relationships from this graph.…”

Performing Weakly Supervised Retail Instance Segmentation via Region Normalization

“…Towards the weakly supervised instance segmentation task [11], [12] in the retail industry, we propose to explore the characteristics of retail industry items and make specific designs to improve the segmentation performance. Because the 1 For example, the MVTec D2S dataset [10] contains four kinds of apples with different prices, i.e.…”