Occluded Human Instance Segmentation

Student: Panaetov Aleksandr

Educational Programme: Applied Mathematics and Information Science (Bachelor)

Year of Graduation: 2020

Standard instance segmentation models based on Mask R-CNN detect all objects and then segment objects individually in each bounding-box. This approach allows to get good segmentation quality in simple cases but it is very complicated to distinguish people with the same bounding-box so models based on detection usually have poor results on images with occlusions like hugging or sports activities. We research human instance segmentation methods which do not rely on detection and explore in more detail Pose2Seg approach which uses human keypoints instead of bounding-box. We suggest some improvements to Pose2Seg architecture and we get significantly better results on the challenging OCHuman dataset which consists of heavily occluded and complex scenes.

Student Theses at HSE must be completed in accordance with the University Rules and regulations specified by each educational programme.

Summaries of all theses must be published and made freely available on the HSE website.

The full text of a thesis can be published in open access on the HSE website only if the authoring student (copyright holder) agrees, or, if the thesis was written by a team of students, if all the co-authors (copyright holders) agree. After a thesis is published on the HSE website, it obtains the status of an online publication.

Student theses are objects of copyright and their use is subject to limitations in accordance with the Russian Federation’s law on intellectual property.

In the event that a thesis is quoted or otherwise used, reference to the author’s name and the source of quotation is required.

Search all student theses