Mediar-V is a multi-head extension of [MEDIAR-Former](https://github.com/Lee-Gihun/MEDIAR) for instance segmentation of overlapping cell structures in microscopy images (e.g., cytoplasm and nucleus of the same cell). Classical flow-field based methods typically support only a single object class and therefore require a separate model for each class. Mediar-V keeps a single backbone and flow representation, but adds dedicated segmentation heads for every target class and trains them jointly, which yields richer supervision and feature sharing between object types. On phase-contrast Glioma C6 and histological CytoNuke datasets with multiple cell classes, this unified architecture outperforms a set of separate single-class models by ≈3 p.p. F1 and ≈4 p.p. AP on average, while using a shared post-processing pipeline that reduces memory usage and FLOPs.
This repository provides two main scripts to configure and run a cell segmentation workflow:
This repository provides two main scripts to configure and run a cell segmentation workflow:
* **generate\_config.py**: Interactive script to create JSON configuration files for training or prediction.
* **generate_config.py**: Interactive script to create JSON configuration files for training or prediction.
* **main.py**: Entry point to train, test, or predict using the generated configuration.
* **main.py**: Entry point to train, test, or predict using the generated configuration.
> Unlike prediction testing, it is not necessary that the specified test directory contains a folder with true masks.
> Unlike prediction testing, it is not necessary that the specified test directory contains a folder with true masks.
### Run multiple configs from Python
You are not limited to CLI arguments: if you have many configs, you can specify them directly in `main.py` (or another script) and call `main()` in manual mode:
