BBAM: Bounding Box Attribution Map for Weakly Supervised Semantic and Instance Segmentation

Lee, Jungbeom; Yi, Jihun; Shin, Chaehun; Yoon, Sungroh

doi:10.48550/arxiv.2103.08907

Cited by 2 publications

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Precisely, under the supervision of candidate masks, the segmentation network is first trained and then choose the better masks for the next training repetition. [60] utilized a Simple Does It (SDI) which is a repetitive training method to slowly enhance generated label estimates. But SDI uses a GrabCut like algorithm for the very first lable estimate creation, BoxSup uses an unsupervised area proposal approach like Multiscale Combinatorial Grouping (MCG) [59].…”

Section: Methods Based On Bounding-box-level Supervisionmentioning

confidence: 99%

“…SDI leaves the training algorithm unchanged and concentrates on externally removing noisey input labels through utilizing prior knowledge. [60] obtained the pseudo mask by applying Grabcut to bounding boxes. Not only using CRF [62] introduced the boxdriven class wise masking mode (BCM), with the filling rate guided adaptive loss (FR0Loss), trying to get rid of the incorrectly annotated regions in the pseudo mask.…”

Section: Methods Based On Bounding-box-level Supervisionmentioning

confidence: 99%

See 1 more Smart Citation

A Brief Survey on Weakly Supervised Semantic Segmentation

et al. 2022

View full text Add to dashboard Cite

Semantic Segmentation is the process of assigning a label to every pixel in the image that share same semantic properties and stays a challenging task in computer vision. In recent years, and due to the large availability of training data the performance of semantic segmentation has been greatly improved by using deep learning techniques. A large number of novel methods have been proposed. However, in some crucial fields we can't assure sufficient data to learn a deep model and achieves high accuracy. This paper aims to provide a brief survey of research efforts on deep-learning-based semantic segmentation methods on limited labeled data and focus our survey on weakly-supervised methods. This survey is expected to familiarize readers with the progress and challenges of weakly supervised semantic segmentation research in the deep learning era and present several valuable growing research points in this field.

show abstract

Section: Methods Based On Bounding-box-level Supervisionmentioning

confidence: 99%

Section: Methods Based On Bounding-box-level Supervisionmentioning

confidence: 99%