LibreRealESRGANx4-restore

RRDBNet Real-ESRGAN super-resolution (4x upscaling), repackaged for LibreYOLO.

Source

Weights derived from xinntao/Real-ESRGAN release RealESRGAN_x4plus.pth. Copyright (c) 2021 Xintao Wang. Licensed under the BSD 3-Clause License.

The RRDBNet architecture lineage is XPixelGroup/BasicSR (Apache-2.0, Copyright 2018-2022 BasicSR Authors).

Modifications

State-dict metadata-wrap only: keys and learned parameters are unchanged; the checkpoint is wrapped in the LibreYOLO v1.0 schema (task=restore, scale=4). Tensor-level parity vs the official model is exact (max_abs_diff == 0, fp32). See weights/convert_realesrgan_weights.py in the LibreYOLO source repository.

Usage

from libreyolo import LibreYOLO

model = LibreYOLO("LibreRealESRGANx4-restore.pt")
res = model.predict("small.jpg")     # res[0].restored is 4x the input
res[0].save("upscaled.png")
# large images: model.predict("big.jpg", tile=512)   # seam-free tiling

License

BSD 3-Clause (weights) with Apache-2.0 architecture lineage. See the LICENSE and NOTICE files in this repository.