Leaderboard 1: 3D denoising (Planaria)¶

👉 Download the training data here - Data on Zenodo

General information¶

This leaderboard is dedicated to denoising 3D fluorescence microscopy images of S. mediterranea, a freshwater flatworm that’s a popular model in regenerative biology and stem cell research.

The data comes from the Planaria denoising dataset originally published in Weigert, M., Schmidt, U., Boothe, T. et al. Content-aware image restoration: pushing the limits of fluorescence microscopy. Nat Methods 15, 1090–1097 (2018).

Planaria were imaged using spinning disk confocal microscopy, under four different combinations of laser power and exposure time. These imaging setups produced 3D image stacks with different levels of noise, simulating realistic low-light constraints often encountered in microscopy imaging.

To prepare the dataset for supervised training, the raw large 3D volumes were cropped into stacks of different sizes. For each sample, we provide:

• A noisy input volume (from one of three lower-exposure conditions)
• A corresponding ground truth (GT) volume (from the highest-quality acquisition using higher laser power and longer exposure).

Training Dataset¶

The training dataset contains noisy-GT image pairs. It is organized into two folders, based on crop size:

├── big_crops
│   ├── gt
│   └── noisy
└── small_crops
  ├── gt
  └── noisy

These are local patches sampled from various planaria volumes.
Contains 17,900 3D stacks of shape (16, 64, 64).

These represent larger regions of the full imaging volumes and can provide larger structural context for training.
Contains 45 3D stacks of average shape (95, 1024, 1024).

In both folders:

• Noisy inputs are stored in the noisy/ subfolder
• Corresponding high-SNR targets are in the gt/ subfolder
• File names are aligned across folders (e.g., noisy/00001.tif ↔ gt/00001.tif).

Evaluation¶

For evaluation, we’ll use a held-out set of 3D planaria stacks that weren't included in the training data. These samples were collected under similar imaging conditions.

Your goal is to build a model that generalizes well across varying noise levels.

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

✍️ Submit here - Submission page on GC