Home

Awesome

Instance Semantic Segmentation List

This repository contains lists of state-or-art instance semantic segmentation works. Papers and resources are listed below according to method types.

Brief introduction

Instance semantic segmentation is a area closely related to detection and semantic segmentation. In particular, it could be seen as detection plus foreground mask. But mostly it is not able to segment non-object pixels such as sky, land etc(which considered as a scene parsing task under semantic segmentation). For quick review related topics, see these survey papers:

Speed/accuracy trade-offs for modern convolutional object detectors, CVPR 2017

Survey of recent progress in semantic image segmentation with CNNs, until 201706

Dataset and benchmark here

DatasetTrainValLinkNote
Pascal VOC 12 Aug105821449SegVOC12origin train 1464+SDB
Pascal VOC SDB val56235732SDBsimilar to VOC12 Main
COCO115k5kCOCOcoco_2014_minival
CityScapes5000/CityScapesevaluation server

Note: Pascal VOC could refer to different split of dataset. Original VOC12 segmentation task consists of train/val/test 1464/1449/1456 images respectively without instance information. It is designed for semantic segmentation. SDB provides instance-aware annotations for images from Pascal VOC12. And their split(8k/3k) differ from VOC, so another split from them Pasval VOC SDB val is provided, which is similar to Pascal VOC Main split(5717/5823).

<h2 id="1">1.Detection-based methods</h2> <h2 id="2">2.Segmentation-based methods</h2> <h2 id="3">3.Others</h2>