DeepFH segmentations for superpixel-based object proposal refinement

作者：

Highlights：

• We propose a superpixel-based refinement system for object proposal generation.

• A novel Felzenszwalb-Huttenlocher-style segmentation utilizing deep features.

• Utilizing the deep features leads to clearly improved segmentations.

• Superpixel-refined object proposals adhere significantly better to object boundaries.

• The results clearly outperform other state of the art methods on the COCO dataset.

摘要

Highlights•We propose a superpixel-based refinement system for object proposal generation.•A novel Felzenszwalb-Huttenlocher-style segmentation utilizing deep features.•Utilizing the deep features leads to clearly improved segmentations.•Superpixel-refined object proposals adhere significantly better to object boundaries.•The results clearly outperform other state of the art methods on the COCO dataset.

论文关键词：Object proposals,Image segmentation,Superpixels

论文评审过程：Received 28 June 2021, Accepted 26 July 2021, Available online 31 July 2021, Version of Record 19 August 2021.

论文官网地址：https://doi.org/10.1016/j.imavis.2021.104263