48 acquisitions were collected across seven ovine kidney sets (K1-K7) and three liver sets (L1-L3) under laparoscopic and surgical overhead lighting, following hemispherical-sweep and trocar (RCM) trajectories. Each acquisition provides 16-band snapshot HSI cubes (XIMEA MQ022HG-IM-SM4X4-VIS) and RGB frames, together with hand-eye-calibrated camera poses derived from a robotic arm end-effector, radiometric white/dark references, and pre-computed RGB-HSI registration homographies.

The dataset is the training and evaluation benchmark for HyKey (Saikia et al., IJCARS 2026), where HSI-based keypoint detection outperformed HSI and RGB baselines, achieving 96.62% mean matching accuracy and 67.18% mAA@10° on pose estimation.

Contents per acquisition: raw HSI mosaics (.npy), raw RGB frames (.npy), hsi_poses.csv, rgb_poses.csv, frame_log.csv, radiometric references, lighting and trajectory info.

License: CC BY-NC 4.0. For commercial use, contact the authors.

Code & model weights & Dataset Description: https://github.com/alexsaikia/HyKey-Hyperspectral-Keypoint-Detection

If you use this dataset please cite the following articles:
Paper for HyKey: https://link.springer.com/article/10.1007/s11548-026-03633-z

Paper for Robotic Arm Acquisition Platform: https://ieeexplore.ieee.org/abstract/document/10879583