Towards Good Practices for Instance Segmentation

TL;DR

This paper explores effective refinement techniques for the Hybrid Task Cascade network to improve instance segmentation performance, achieving notable results on the COCO datasets.

Contribution

It systematically evaluates various refinements to HTC and demonstrates their impact on segmentation accuracy through comprehensive ablation studies.

Findings

Achieved 0.47 mAP on COCO test-dev

Achieved 0.47 mAP on COCO test-challenge

Provided insights into refinement effects on model performance

Abstract

Instance Segmentation is an interesting yet challenging task in computer vision. In this paper, we conduct a series of refinements with the Hybrid Task Cascade (HTC) Network, and empirically evaluate their impact on the final model performance through ablation studies. By taking all the refinements, we achieve 0.47 on the COCO test-dev dataset and 0.47 on the COCO test-challenge dataset.