Mat2Seg: An RGB-NIR Dataset and Benchmark for Material Segmentation of Household Objects

RGB-NIR material segmentation for household objects.

Household objects shown in paired RGB and near-infrared views

Overview

Mat2Seg is an RGB-NIR dataset and benchmark for material segmentation of household objects. It combines visible and near-infrared imagery to study how complementary sensing can improve the delineation of material regions.

The benchmark focuses on material boundaries within everyday object scenes, supporting the evaluation of methods that reason about materials at the pixel level rather than assigning a single label to an entire image.

Paired RGB-NIR Capture

RGB

Visible-spectrum color, texture, and appearance cues from everyday object scenes.

NIR

Near-infrared observations that offer complementary signatures for distinguishing material regions.

Benchmark

01

Material regions

Evaluate pixel-level delineation of different materials within a shared household scene.

02

Multimodal comparison

Compare RGB-only, NIR-only, and fused RGB-NIR approaches under a common segmentation task.

03

Household contexts

Study material boundaries across familiar objects, surfaces, and mixed-material arrangements.

Release

Mat2Seg release package

Access the Mat2Seg dataset, annotations, and benchmark materials.

Citation

Citation details will be published with the official Mat2Seg release.

@misc{mat2seg,
  title = {Mat2Seg: An RGB-NIR Dataset and Benchmark for Material Segmentation of Household Objects},
  note = {Dataset release in preparation}
}