Published August 31, 2026
| Version v1
Model
Open
SWITi Model Checkpoints: Demonstrating Tiling Artifact Attenuation with Sliding Window Inner Tiling
Authors/Creators
-
1.
Human Technopole
-
2.
Università Campus Bio-Medico
Description
This record contains MicroSplit model checkpoints that were used to demonstrate the attenuation of tiling artifacts when using sliding-window-inner-tiling (SWITi) in comparison to inner-tiling with a Minimum Mean Squared Error (MMSE) estimate. There are 3 sets of checkpoints that were trained on three different datasets, PaviaATN, HT-LIF24, and CBG-Z18; the training, validation and test splits are available on Zenodo. The checkpoints exhibit tiling artifacts in their predictions of the test splits. These exact checkpoints were used to report results in papers introducing the methods MicroSplit for CBG-Z18 and HT-LIF24 and scSplit for PaviaATN.
There are 3 directories, named after each dataset the models were trained on. Contained in each directory are the best and last checkpoints, BaselineVAECL_best.ckpt and BaselineVAECL_last.ckpt respectively. Additionally included is a config.yaml file containing the original training and model parameters for each set of checkpoints.
The checkpoints were trained with the ashesh-0/Disentangle repository.
Links
Code to reproduce results in the SWITi paper is available at: https://github.com/juglab/SWITi_reproducibility.Files
switi_checkpoints.zip
Additional details
Related works
- Is derived from
- Journal article: 10.1038/s41592-026-03082-1 (DOI)
- Conference paper: 10.52202/085713-2160 (DOI)
- Is supplement to
- Preprint: 10.48550/arXiv.2607.18990 (DOI)
Funding
Software
- Repository URL
- https://github.com/juglab/SWITi_reproducibility
- Programming language
- Python
- Development Status
- Concept