Pre-trained weights of Pix2Rep-v2 (MICCAI 2026) pre-training on the AASCE 2019 dataset. Pix2Rep-v2 pre-training provides powerful pixel-level embeddings suitable across dense prediction tasks. These weights can be used as initialization for training Pix2Point (faster convergence and better results on spinal landmark detection).
This model is presented in the paper Dense Representation Learning and Differentiable Shape Modeling for Spine Landmark Detection.
See https://github.com/BioMedTP/Pix2Point for details on how to use the pre-trained model.