
RetinaRegNet, a versatile model, achieves state-of-the-art performance on various retinal image registration tasks through innovative techniques, including diffusion feature extraction, inverse consistency constraint, and a transformation-based outlier detector.


coremsg

versatile-retinal-image-registration-model-retinaregnet-outperforms-state-of-the-art-methods-across-multiple-datasets


Versatile Retinal Image Registration Model: RetinaRegNet Outperforms State-of-the-Art Methods Across Multiple Datasets


title_rewrite


The authors propose to improve the ConKeD framework for retinal image registration by testing multiple contrastive learning loss functions, including SupCon, MP-InfoNCE, MP-N-Pair, and FastAP losses. They demonstrate state-of-the-art performance across multiple datasets, including the standard FIRE benchmark as well as two new datasets (LongDRS and DeepDRiD) with diverse characteristics.


improving-descriptor-learning-for-retinal-image-registration-a-comprehensive-study-of-contrastive-losses


Improving Descriptor Learning for Retinal Image Registration: A Comprehensive Study of Contrastive Losses
