Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
scSplit: Bringing Severity Cognizance to Image Decomposition in Fluorescence Microscopy
Authors: Ashesh Ashesh, Florian Jug
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show that this method solves two relevant tasks in fluorescence microscopy, namely image splitting and bleedthrough removal, and empirically demonstrate the applicability of sc Split on 5 public datasets. The source code with pre-trained models is hosted at https://github.com/juglab/sc Split/. ... 4 Experiments and Results |
| Researcher Affiliation | Academia | Ashesh Ashesh Human Technopole EMAIL Florian Jug Human Technopole EMAIL |
| Pseudocode | No | The paper describes the method sc Split using formal notation and explanations (e.g., in Section 3 'Our Method'), but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code with pre-trained models is hosted at https://github.com/juglab/sc Split/. ... Our code will be publicly available under a permissive license. |
| Open Datasets | Yes | We show that this method solves two relevant tasks in fluorescence microscopy, namely image splitting and bleedthrough removal, and empirically demonstrate the applicability of sc Split on 5 public datasets. ... We tackle five tasks coming from five real microscopy datasets, namely Hagen et al. [29], Bio SR [30], HTT24 [6], HTLIF24 [6], and Pavia ATN [3]. |
| Dataset Splits | No | The paper refers to 'training data' and 'test set' for various datasets (e.g., in Section 4 and Supplementary Section E & J), but it does not provide explicit details on how these datasets were split into training, validation, and test sets, such as specific percentages, sample counts, or a detailed splitting methodology. |
| Hardware Specification | Yes | training (taking 3 days on a Tesla V100 GPU). ... On our highperformance cluster (Intel(R) Xeon(R) Gold 5220 CPU @ 2.20GHz), it takes around 28 milliseconds (measured with %timeit in ipython) to run torch.nn.Instance Norm2d(1) on a 3000 3000 image. |
| Software Dependencies | No | The paper mentions 'torch.nn.Instance Norm2d(1)' in Supplementary Section B, implying the use of PyTorch, and 'ipython' for measurement. However, it does not provide specific version numbers for these or any other key software components, which is required for a reproducible description of ancillary software. |
| Experiment Setup | Yes | We use a patch size of 512 to train sc Split, In DI, U-Net and denoi Split networks. ... For training Geni networks, we use MAE loss and for training Reg network, we use MSE loss. We use Adam optimizer with a learning rate of 1e-3. ... For a training schedule of 450K iterations with a batch size 8 (our configuration). |