RGB-D Individual Segmentation

TL;DR

This paper introduces the CoLA pipeline for fine-grained individual segmentation using RGB-D data, addressing challenges like limited training data and unknown backgrounds, and demonstrates superior performance on new and existing datasets.

Contribution

The paper proposes a novel 'Context Less-Aware' (CoLA) method for individual segmentation that effectively utilizes RGB-D data and scale-aware training, outperforming baseline methods.

Findings

CoLA significantly improves segmentation accuracy on YCB-Video dataset.

Proposed method outperforms baselines on the new Supermarket-10K dataset.

Code and datasets will be publicly released.

Abstract

Fine-grained recognition task deals with sub-category classification problem, which is important for real-world applications. In this work, we are particularly interested in the segmentation task on the \emph{finest-grained} level, which is specifically named "individual segmentation". In other words, the individual-level category has no sub-category under it. Segmentation problem in the individual level reveals some new properties, limited training data for single individual object, unknown background, and difficulty for the use of depth. To address these new problems, we propose a "Context Less-Aware" (CoLA) pipeline, which produces RGB-D object-predominated images that have less background context, and enables a scale-aware training and testing with 3D information. Extensive experiments show that the proposed CoLA strategy largely outperforms baseline methods on YCB-Video dataset and our proposed Supermarket-10K dataset. Code, trained model and new dataset will be published with this paper.