ASPECT-10

Description

ASPECT-10 (Aggregate Segmentation and Perception Evaluation in Critical Traits) is a field-acquired benchmark for evaluating aggregate boundary detection and instance segmentation under heterogeneous real-world conditions. The dataset contains 30 orthorectified image patches, organized into 10 diagnostic subsets with three spatially distinct observations per subset. The benchmark covers variations in inter-particle contrast, intra-particle appearance, and illumination/radiometric conditions. Ground truth includes approximately 780K labeled boundary pixels and 4,866 verified particle instances. The dataset is organized into three main directories. IMAGES contains the RGB input images. Boundary_GT contains the ground-truth inter-particle boundary annotations. It includes two subfolders: mat and png. The mat folder contains MATLAB .mat files with the boundary ground-truth data, while the png folder contains the corresponding boundary annotation images. Mask_GT contains the instance-level ground-truth annotations. It includes two subfolders: JSON and png. The JSON folder contains .json files with the instance-level annotation data, while the png folder contains the corresponding ground-truth mask images. The corresponding image and subset identifiers are encoded in the filenames. ASPECT-10 subset mapping: G1-S1: Agg_Crushed_White_S_GLC G1-S2: Agg_Rounded_Dust_S_GLC G2-S1: Agg_Crushed_Black_S_GLC G2-S2: Agg_Crushed_Clear_M_GLC G3-S1: Agg_Crushed_Green_M_GLC G3-S2: Agg_Crushed_Green_S_GLC G3-S3: Agg_Crushed_Red_S_GLC G3-S4: LightAgg_Black_M_HTC G4-S1: Agg_Crushed_Black_S_HLC G4-S2: Agg_Rounded_Black_M_HLC

Institutions

• Sharif University of Technology

  Tehran, Tehran

Categories

Licence