v1.2

What's Changed

• Change YouTube link and update video description by @romeokienzler in #1008

• Add diffusers dependency with version constraint by @romeokienzler in #1010

• Terramind integration test by @WanjiruCate in #1013

• Geobench2-visualization by @naomi-simumba in #1021

• update type hint in generic multimodal datamodule by @Isabelle-Wittmann in #1025

• It will be moved to the documentation by @Joao-L-S-Almeida in #1006

• ScalarRegressionTask by @leonardijulia in #929

• Update TerraMind dtypes by @blumenstiel in #1033

• Improvments for multimodal data by @blumenstiel in #1009

• Cleaning up cache in two stages to avoid OSError by @Joao-L-S-Almeida in #1041

• Deprecated classification head warning by @blumenstiel in #1039

• Always generated float16 dummy inputs (vLLM default) by @christian-pinto in #1044

• Updating Discord link to avoid expiration by @Joao-L-S-Almeida in #1045

• fixed issues with the tutorial. missing model and broken file import by @Mdmdma in #1042

• Update/albucore-jsonargparse-pytorch by @romeokienzler in #1040

• Improvements to PixelWiseModel Padding Handling

This update fixes issues in how padding is handled by PixelWiseModel's forward method, by moving automatic padding from bottom right to all corners, with the important caveat that this change is not backward compatible with TerraTorch versions prior to 1.2. This means that models must be re-trained or fine-tuned with TerraTorch ≥ 1.2 for inference, if the model was trained using padding (training image size not divisible by 2 * patch size, i.e. 32 pixels for most models).

For more technical details, see Issue #1079

New Contributors

• @leonardijulia made their first contribution in #929
• @Mdmdma made their first contribution in #1042

Full Changelog: v1.1.1...v1.2