Cellpose Model Training

Docker setup for training custom Cellpose models with GPU support.

Usage

1. Prepare your data:

   data/
   ├── train/
   │   ├── image_001.tif
   │   ├── image_001_masks.tif
   │   ├── image_002.tif
   │   └── image_002_masks.tif
   └── test/
       ├── test_001.tif
       └── test_001_masks.tif
   

2. Build and run:

   docker build -t cellpose-train .
   docker run --gpus all -v ./data:/data -v ./models:/models cellpose-train

3. Custom training parameters:

   docker run --gpus all -v ./data:/data -v ./models:/models cellpose-train \
     python train_cellpose.py --learning_rate 0.0001 --n_epochs 50

Output

• Trained models saved to ./models/
• Training logs available in container output

Performance

• Training time: ~1 hour on RTX 4090 D with default settings (100 epochs, 540 training images)

Requirements

• NVIDIA Docker runtime
• CUDA-compatible GPU