🔬 Leaderboard 4: 2D denoising (FMD)¶

👉 Download the training data here - Data on Zenodo

General Information¶

This leaderboard is based on the Fluorescence Microscopy Denoising (FMD) dataset, introduced by Zhang et al. in A Poisson-Gaussian Denoising Dataset with Real Fluorescence Microscopy Images.

This dataset spans multiple microscopy modalities - confocal, two-photon, and wide-field - and covers diverse biological samples including cells, zebrafish, and mouse brain tissues.

Training Dataset¶

The training data includes noisy ground-truth image pairs collected across multiple conditions. Each sample consists of multiple noisy views of the same field of view, acquired over time, so while the structures remain identical, the noise varies across frames.

We provide filenames with metadata so that you can sort or filter by:

• Imaging modality
• Channel/structure (e.g., nuclei, tubulin, actin)
• Noise level

Sample and timepoint ID Each filename follows this format: {modality}_{channel}_{noiseLevel}_{sampleID}_{timestampID}.tif

For example: CF_R_L1_1_1.tif

...stands for: Confocal microscopy, channel R, noise level 1, sample 1, timestamp 1. All image files are 2D .tif slices of the same size.
Ground truth images were generated by averaging across multiple noisy frames (i.e., temporal average over 50 frames of the same sample).

Evaluation¶

We’ll evaluate submissions on a held-out set of images not included in the training set but following the same acquisition protocol, with a mix of modalities, channels, and noise levels.
You goal is to build a model that can generalize across diverse imaging conditions, including microscopes, biological structures, and noise levels, while preserving fine structural detail.

👀 Check out the example submission here - Example submission on GitHub

✍️ Submit here - Submission page on GC