Author: msiplab

Posted on

Released TanSacNet Pre-release20250105

We updated TanSacNet Project for developing tangent space adaptive control networks.

  • New PyTorch Implementation: Introduced a PyTorch-based 2-D LSUN framework, including batch processing in orthonormalTransform.py and a low-dimensional approximation sample. Optimized gradient computation and sequential matrix processing to improve efficiency in PyTorch and CUDA environments.
  • Enhanced MATLAB Support: Improved data type and device management to ensure stability and performance in MATLAB workflows.
  • Code Refactoring and Stability: Streamlined code structure for maintainability and resolved initialization and CPU processing issues across both frameworks.

This pre-release highlights the new PyTorch implementation, along with key updates to MATLAB support and overall code stability.

Paper Published in IAPSIPA Transactions on Signal and Information Processing

 

Posted on

Paper Published in ITE Transactions on Media Technology and Applications

The following paper has been published in ITE Transactions on Media Technology and Applications.

  • Jikai Li, Shogo Muramatsu, [Paper] Structured Deep Image Prior for Image Denoising with Interscale SURE-LET, ITE Transactions on Media Technology and Applications, 2025, Volume 13, Issue 1, Pages 187-199, Released on J-STAGE January 01, 2025, Online ISSN 2186-7364, https://doi.org/10.3169/mta.13.187, https://www.jstage.jst.go.jp/article/mta/13/1/13_187/_article/-char/en,
  • Abstract:
    This study develops a self-supervised image denoising technique that incorporates a structured deep image prior (DIP) approach with Stein’s unbiased risk estimator and linear expansion of thresholding (SURE-LET). Leveraging interscale and interchannel dependencies of images to develop a multichannel denoising approach. The original DIP, introduced by Ulyanov et al. in 2018, requires a random image as the input for restoration, offering an advantage of not requesting training data. However, the interpretability of the role of the network is limited, and challenges exist in customizing its architecture to incorporate domain knowledge. This work integrates SURE-LET with Monte Carlo computation into the DIP framework, providing the reason of the random image supply and shifting the focus from generator to restorer design, thus enabling the network structure of DIP to more easily reflect domain knowledge. The significance of the developed method is confirmed through denoising simulations using the Kodak image dataset.

 

Posted on

Paper Published in IAPSIPA Transactions on Signal and Information Processing

The following paper has been published in APSIPA Transactions on Signal and Information Processing (Q1: Information Systems).

It is a result of Grant-in-Aid for Scientific Research (A) “(JP22H00512) Locally Structured Unitary Networks and Tangent Space Learning for Dynamic Systems Modeling”.

It is Open Access. The source code is available at Code Ocean.

SCImago Journal & Country Rank