CARL: Camera-Agnostic Representation Learning for Spectral Image Analysis

This Zenodo record provides the pretrained model weights for CARL, a camera-agnostic spectral imaging model designed for general-purpose feature extraction. CARL was introduced in our ICLR 2026 paper and the weights released here correspond to the models used in that work. The weights result from large-scale self-supervised pretraining on remote sensing imagery, producing representations that transfer across sensors/cameras and downstream tasks. This record is intended to support reproducibility and to enable researchers and practitioners to use CARL as a pretrained backbone for spectral/remote-sensing applications.