/beta-segmentinator

Primary LanguagePython

UnMicst Segmentation step (Segmentinator... inator)

This repo contains the segmentation step for debugging purposes using Mask-RCNN models.

How to run

  1. Install required packages pip install requirements.txt (advisable to use a virtual environment)
  2. python main.py input output Where input is the tiff file to segment and output is the directory to store the output tff.

The reference model is located at hits/lsp-analysis/UnMICSTdev/FOR ALEX HUMAN ANNOTATIONS USE THESE/models_that_work/model_trained_without_ignore_manyWindowsAndSynthetic.pt

If you find bugs or comments please open an issue in this repo so we can keep track of them, remember to add as much detail as posible.

IMPORTANT

  • This model has been trained with 0.325 microns per pixel, you should be wary of different resolutions.
  • Use gpu, otherwise it take a while.
  • The ---mode-path flag is necesary with the full path to the saved model file to load that instead of the default.
  • For documentation on the available flags, just run python main.py without arguments or python main.py -h.
  • Paramenters can be set with the --thres-* flags like --thres-nms.
  • Default running device is GPU 1, if you get error make sure you have a GPU and that it is available to run.