Framework for training spotMAX AI

[ElpadoCan]

GUI

• Select Train Positions
• Select Val Positions
• Select Test Positions
• Select coords table --> generate masks ...
• ... or select masks file
• Spot size for generating masks from coords table
• Select architecture (large, medium, small - 2D or 3D)
• Start from pre-trained --> load weights file
• Activate data augmentation
• Crops shape (default 256,256)
• Folder path for configuration file
• Filename for configuration file
• XY pixel size
• Load from saved workflow
• Enable random shuffling train-val
• Name config yaml file as the workflow file
• Maximum number of crops

Implementation

• Generate masks from positions and coords table
• Generate h5 files with 'X', 'pixel_size', 'y' keys
• Generate ini file training workflow
• Data augmentation: DoG filter, Gaussian filter (2 sigmas)
• Allow relative paths for weights file path in config yaml file

Inference

• Allow loading custom model
• Custom model should be a folder with weights and config yaml file

Notes

• Propose to share data for training on denbi
• Example notebook how to train UNet 2D and 3D
• Documentation

[github-actions]

Hi there, I'm the SpotMAX bot 👋. This issue had no activity for more than 180 days. For now, we have marked it as "inactive" until there is some new activity. If this issue was not solved yet, unfortunately we haven't had the time to implement it or it requires more discussion. Note that it doesn't mean it has been ignored, but a little reminder from your side would help :D. Feel free to reach out to us here or on our forum. If you think that this issue is no longer relevant, you may close it by yourself. In any case, we apologise for the inconvenience and we thank you for your patience and contributions so far!