mhmdjouni/uafl-checkpoints

UAFL Released Checkpoint

This repository mirrors the released checkpoint for Enhancing Unregistered Hyperspectral Image Super-Resolution via Unmixing-based Abundance Fusion Learning (UAFL) by Zhang et al.

The checkpoint corresponds to the released REAL-dataset scale-factor-4 model. This Hugging Face repository is maintained to provide stable and convenient programmatic access to the released artifact.

Checkpoint

sf4/
  ckpt/
    model-best.pth
    model-best.original-torchlight.pth

• model-best.original-torchlight.pth is the untouched checkpoint distributed by the original authors;
• model-best.pth is a Torchlight-free re-serialization intended for ordinary modern PyTorch loading.

All scientific credit for UAFL and the original checkpoint goes to the original authors.

Torchlight-free checkpoint

The original released checkpoint contains serialized historical Torchlight training metadata.

The UAFL model tensors themselves do not depend on Torchlight, but this metadata prevents the original artifact from being loaded with ordinary modern PyTorch when Torchlight is not installed.

The canonical model-best.pth preserves:

checkpoint["module"]["model"]
checkpoint["epoch"]

The epoch field is retained because the released REAL checkpoint corresponds to epoch 300 and the reproduction pipeline explicitly verifies this metadata.

The original and sanitized model states were verified to have:

• identical keys;
• identical tensor shapes;
• identical tensor dtypes;
• tensor-for-tensor equality using torch.equal;
• identical released epoch metadata (epoch == 300).

The sanitized checkpoint was additionally strict-loaded into the released UAFL architecture.

Reproducing the sanitized checkpoint

The repository includes sanitize_torchlight_checkpoint.py.

python sanitize_torchlight_checkpoint.py \
    sf4/ckpt/model-best.original-torchlight.pth \
    sf4/ckpt/model-best.pth \
    --preserve-epoch

The script verifies both tensor equality and preservation of the epoch field.

Loading

The canonical checkpoint can be loaded using ordinary PyTorch:

checkpoint = torch.load("sf4/ckpt/model-best.pth", map_location="cpu")
state_dict = checkpoint["module"]["model"]
model.load_state_dict(state_dict, strict=True)

No Torchlight installation or compatibility loader is required for the canonical checkpoint.

Original work

• Paper: Enhancing Unregistered Hyperspectral Image Super-Resolution via Unmixing-based Abundance Fusion Learning, Yingkai Zhang, Tao Zhang, Jing Nie, and Ying Fu.
• Official implementation: https://github.com/yingkai-zhang/UAFL
• Dataset: https://huggingface.co/datasets/ZeqiangLai/HSI-RefSR

This repository provides the released artifact for convenient automated downloading together with a Torchlight-free copy and the exact sanitation procedure used to produce it.

All scientific credit goes to the original authors.