danaroth/adapters

Description

This repository contains a series of weights for adapting the DRUNet denoiser in order to be able to work with hyperspectral images.

These weights are meant to be used with the hypnp library:

http://github.com/Danaroth83/hypnp

In particular the weights contained in this folder are associated to the following adapting architecture:

• 20250922_160658_495381: Progressive injection with skip attention module.
• 20250922_173031_803562: An encoding/decoder network.
• 20250926_125509_133496: A skip connection with an attention module.
• 20251019_163957_302293: Spectral projection encoder/decoder, the encoder is computed with QR decomposition, while the decoder is a CNN network.
• 20251019_163951_000174: A FiLM that hooks middle layers of the DRUNet. Baseline result.
• 20251019_163950_994277: A FiLM network. With respect to baseline, context from current computation is not passed to the hooks.
• 20251019_190756_177008: A FiLM network. With respect to baseline, this network hooks to all the layers of DRUNet.

Credits

These weights were produced by:

Daniele Picone
Univ. Grenoble Alpes, CNRS, Grenoble INP, GIPSA-lab, 38000 Grenoble, France
Mail: [email protected]