Learning Pixel Representations for Generic Segmentation

TL;DR

This paper introduces Deep Generic Segmentation (DGS), a novel deep learning method that learns pixel-wise representations for non-semantic segmentation, improving quality and achieving state-of-the-art segment similarity scores.

Contribution

It presents a new approach for learning pixel representations for generic segmentation, moving beyond edge detection and aligning with semantic segmentation techniques.

Findings

Learned meaningful pixel representations that improve segmentation quality.

Achieved state-of-the-art segment similarity scores.

Results are competitive and promising.

Abstract

Deep learning approaches to generic (non-semantic) segmentation have so far been indirect and relied on edge detection. This is in contrast to semantic segmentation, where DNNs are applied directly. We propose an alternative approach called Deep Generic Segmentation (DGS) and try to follow the path used for semantic segmentation. Our main contribution is a new method for learning a pixel-wise representation that reflects segment relatedness. This representation is combined with a CRF to yield the segmentation algorithm. We show that we are able to learn meaningful representations that improve segmentation quality and that the representations themselves achieve state-of-the-art segment similarity scores. The segmentation results are competitive and promising.