Fine-Grained Object Counting
Size: 1120 images | Year: 2025 | Status: Available
Counting specific objects when there are multiple visually similar objects in the scene is a challenging task. LookAlikes contains 14 broad categories and 37 fine-grained categories and is designed to benchmark fine-grained object counting. The dataset is available now and can be downloaded from the dataset page.
Counting Stacked Objects (Multi-View / 3D)
Size: 13,100+ scenes | Year: 2025 | Status: Released
Counting objects piled into a 3D stack requires reasoning about the objects you cannot see. StackCounting pairs real multi-view scenes with ground-truth counts and geometry (Stacks-3D-Real), a large set of synthetic 3D stacks with renders and meshes (Stacks-3D-Synth), and single-view real scenes with visible-object counts (Stacks-2D-Real). It accompanies Counting Stacked Objects (ICCV 2025, Oral) by Dumery et al.; I contributed a large portion of the real-world data capture.