README.md

@@ -1,19 +1,16 @@
-# Mediar-V
+# Cell Segmentator
 
 ---
 
 ## Overview
 
-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:
 
-* **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.
 
 ---
 
-
 ## Installation
 
 0. **Install uv**:
@@ -220,20 +217,6 @@ python main.py -c config/templates/predict/YourConfig.json -m predict
 
 > 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:
-
-```python
-from main import main
-
-for cfg in [
-    "config/templates/train/YourConfigA.json",
-    "config/templates/train/YourConfigB.json",
-]:
-    main(manual=True, config_path=cfg, mode="train")
-```
-
 ---
 
 ## Acknowledgments

core/models/__init__.py

@@ -2,13 +2,13 @@ from torch import nn
 from typing import Final, Type, Any
 from pydantic import BaseModel
 
-from .mediar_v import MediarV, MediarVParams
+from .model_v import ModelV, ModelVParams
 
 
 __all__ = [
     "ModelRegistry",
-    "MediarV",
-    "MediarVParams"
+    "ModelV",
+    "ModelVParams"
 ]
 
 
@@ -17,9 +17,9 @@ class ModelRegistry:
     
     # Single dictionary storing both model classes and parameter classes.
     __MODELS: Final[dict[str, dict[str, Type[Any]]]] = {
-        "MediarV": {
-            "class": MediarV,
-            "params": MediarVParams,
+        "ModelV": {
+            "class": ModelV,
+            "params": ModelVParams,
         },
     }
     

core/models/model_v.py

@@ -7,10 +7,10 @@ from segmentation_models_pytorch.base.modules import Activation
 
 from pydantic import BaseModel, ConfigDict
 
-__all__ = ["MediarV"]
+__all__ = ["ModelV"]
 
 
-class MediarVParams(BaseModel):
+class ModelVParams(BaseModel):
     model_config = ConfigDict(frozen=True)
     
     encoder_name: str = "mit_b5"                               # Default encoder
@@ -26,19 +26,19 @@ class MediarVParams(BaseModel):
     
     def asdict(self) -> dict[str, Any]:
         """
-        Returns a dictionary of valid parameters for `nn.MediarV`.
+        Returns a dictionary of valid parameters for `nn.ModelV`.
 
         Returns:
-            dict(str, Any): Dictionary of parameters for nn.MediarV.
+            dict(str, Any): Dictionary of parameters for nn.ModelV.
         """
         loss_kwargs = self.model_dump()
         return {k: v for k, v in loss_kwargs.items() if v is not None}  # Remove None values
 
 
-class MediarV(MAnet):
-    """MediarV model"""
+class ModelV(MAnet):
+    """ModelV model"""
 
-    def __init__(self, params: MediarVParams) -> None:
+    def __init__(self, params: ModelVParams) -> None:
         # Initialize the MAnet model with provided parameters
         super().__init__(**params.asdict())