CellFMCount model checkpoints (CellFMCount / SAM-Counter study)

This release provides the pretrained model checkpoints used in the CellFMCount paper. Each checkpoint corresponds to the model with the lowest validation MAE (Mean Absolute Error), which measures the average absolute difference between predicted and true cell counts per image. See the repository for training, inference and evaluation scripts. Models included span four families:Regression-based: VGG-16, ResNet-18, ResNet-50, EfficientNet-B7Crowd-counting density-map estimation (DME): MCNN, CSRNetCell-specific DME: SAU-Net, C-FCRN+AUX, Count-ception, FCRN-ASAM-based: SAM-CounterSAM-Counter adapts a Segment Anything Model (SAM) ViT encoder and adds lightweight convolutional layers for density-map estimation, enabling cell counting with strong generalization. The overall purpose of the project is to provide a standardized benchmark and strong baseline models for fluorescence microscopy cell counting, enabling reproducible comparison of methods. By releasing both the dataset and model checkpoints, the work aims to accelerate the development and adoption of robust, data-driven cell counting approaches in biological and biomedical research.