DARE 2D

Creation (v1): 27/10/2025

Update (v2) 27/07/2026 Adds pytorch models

This repository contains the data and pretrained models associated with the paper:

Romain Karpinski†, Alice Gros◊, Marc Karnat*, Qazi Saaheelur Rahaman*◊, Jules Vanaret◊, Mehdi Saadaoui◊, Sham Tlili◊, Jean-François Rupprecht*
* Aix Marseille Univ, CNRS, LAI (UMR 7333), Turing Centre for Living systems, Marseille, France
† LORIA, CNRS, Nancy, France
◊ Aix Marseille Univ, CNRS, IBDM (UMR 7288), Turing Centre for Living systems, Marseille, France

Contact: romain.karpinski@loria.fr, sham.tlili@univ-amu.fr, jean-francois.rupprecht@univ-amu.fr; 

—

## Description

We propose a two-stage deep learning method to characterize cell divisions in time-lapse microscopy sequences.

1. **Stage 1 – Division Detection:**  
   A semantic segmentation network (U-Net) identifies potential division events within image sequences.

2. **Stage 2 – Regression of Division Geometry:**  
   A convolutional regression model estimates the orientation and separation of daughter cells.

The method was applied to confocal image sequences of neural tube formation in chicken embryos.  
Optimization of the networks was performed through systematic hyperparameter exploration.

—

## Repository contents

– **data.zip:** Dataset used for training and testing.  
– **model.zip:** Pretrained weights, tf and pytorch.

—

## Citation

If you use this dataset or pretrained models, please cite:

Romain Karpinski, Alice Gros, Marc Karnat, Qazi Saaheelur Rahaman, Jules Vanaret, Mehdi Saadaoui, Sham Tlili, and Jean-François Rupprecht, “DARE: Division Axis REcognition from time-lapse image sequences in 2D and 3D” (bioRxiv 2026). 

—

## License

This work is distributed under the **CC BY 4.0 License**.  
You are free to use, share, and adapt with attribution.