Revisiting the Encoding of Satellite Image Time Series

The model weights for Exchanger+Unet and Exchanger+Mask2Former, accompanying the BMVC 2023 paper titled "Revisiting the Encoding of Satellite Image Time Series". The code has been made publicly available at the link. The accompanying video and slides can be found through the link for a better intuitive understanding. This version provides the weights for Exchanger trained on PASTIS with pixel-set format data, which is used for initializing the downstream dense prediction models. You can check the previous version for weights for dense prediction models.