OpenSR SRGAN pretrained models

This repository packages the supported Sen2 SR release and two maintained legacy presets for the OpenSR SRGAN dev package.

Preset	Model card	Input / scale	Checkpoints
`SEN2-SR`	Sen2 SR v1	B04/B03/B02/B08, 4×	inference, training
`RGB-NIR`	RGB-NIR v1	4 bands, 4×	inference
`SWIR`	SWIR v1	6 bands, 8×	inference

Install the package from the supported branch and use a preset:

python -m pip install 'opensr-srgan @ git+https://github.com/ESAOpenSR/SRGAN.git@dev'

from opensr_srgan import load_inference_model

model = load_inference_model("SEN2-SR", map_location="cpu")
sr_dn = model.predict_step(raw_dn_tensor)  # BCHW raw DN; returns CPU tensors

See examples for tensor, GeoTIFF, and resume-training entry points. Sample artifacts are in data/examples.

Repository layout

models/
  sen2-sr/v1/
    config.yaml
    weights/{inference.ckpt,training.ckpt}
    training/{config.yaml,splits/crosssensor.json}
    training/provenance/{environment.txt,executed_training.yaml,lineage.json,source_at_stop.tar.gz}
    metrics/validation.json
    verification/{inference.json,resume_original.json,resume_promotion.json}
  rgb-nir/v1/{config.yaml,weights/inference.ckpt}
  swir/v1/{config.yaml,weights/inference.ckpt}
data/examples/{rgb-nir_batch.pt,dataset.zip}
examples/{infer_tensor.py,infer_geotiff.py,resume_training.py,README.md}

Each model directory includes a manifest.json and SHA256SUMS for release integrity. The former root-level artifact names remain compatibility aliases for existing downloads; new integrations and documentation use the canonical paths above.

Model weights, model cards, and authored examples are released under the MIT License. See NOTICE.md for the scope of that license and the treatment of archived source and sample data.

Downloads last month: -; Downloads are not tracked for this model. How to track