Dataset and Benchmark for Domain-Agnostic Image-Based Rigid Slice-to-Volume Registration
“Needles & Haystacks” introduces a public benchmark for rigid slice-to-volume registration. The benchmark studies the problem of locating a 2D image slice inside a 3D volume without relying on domain-specific assumptions such as medical landmarks, standard anatomical orientations or segmentation masks.
The work provides a large collection of registration tasks, an online benchmark and baseline methods. It also introduces LoFTR-S2V, a learned detector-free approach for matching 2D slices with 3D volumes.
Find out more:
- Project page: https://xaf-cv.github.io/nh-rs2v/
- Dataset: https://github.com/xaf-cv/NH-RS2V-dataset
- Baselines: https://github.com/xaf-cv/NH-RS2V-baselines
Citation:
A. Frolov, F. Kleiner, C. Rößler, and V. Rodehorst, ‘Needles & Haystacks: Dataset and Benchmark for Domain-Agnostic Image-Based Rigid Slice-to-Volume Registration’, pp. 7081–7091, Feb. 2025.