UnConKeD companion data: artery/vein vessel maps and logits, Messidor-2 FOV masks, crossing/bifurcation keypoints and descriptor network

Companion data for "Unsupervised training of keypoint-agnostic descriptors for flexible retinal image registration", Rivas-Villar, Hervella, Rouco and Novo, Pattern Recognition Letters 208 (2026) 41-47, doi:10.1016/j.patrec.2026.07.015. Code: https://github.com/davidrrvv/unconked-retinal-registration Contents:- av_probs.zip (11.9 GB download, unpacks to 268 files, 18.2 GB) - artery/vein vessel segmentation probabilities (sigmoid output), one .npy per FIRE image, 2912x2912 float64.- av_logits.zip (11.9 GB download, unpacks to 268 files, 18.2 GB) - the same network's raw pre-activation logits. - messidor2_masks.zip (13.0 MB download, unpacks to 1748 files, 21 MB) - Messidor-2 field-of-view masks, binary PNG at each image's own resolution.- cb_keypoints.tar.gz (75 KB download, unpacks to 536 files, 273 KB) - crossing/bifurcation keypoint locations for 268 images.- descriptor_net.pth (5.8 MB) - the trained descriptor network used for the published results. Not included, and must be obtained from their original sources: the FIRE dataset (images, control points and the region-of-interest mask) and the Messidor-2 images. The vessel segmentation network that produced the AV maps is from separate work and is not included. This archive is the authors' own derived work and is licensed CC BY-NC-SA 4.0. That licence does not extend to the third-party datasets these files were derived from.