This project aims to estimate the number of fluorescent fibres present in a UV-illuminated slide image.
After installing conda:
$ git clone https://github.com/hughfdjackson/fluorescent-fibre-counting.git $ cd fluorescent-fibre-counting $ conda env create --file environment $ source activate fluorescent-fibre-counting
Counting fluorescent fibres involves:
- Taking an image of your slide under UV light
- Preprocessing the images
- Feeding the preprocessed images to the model
Taking images under UV light
Refer to the "taking photos of slides" guide for more details.
Once you've gotten your images (preferably in an uncompressed
$ source activate fluorescent-fibre-counting $ python -m preprocess
--max-hue values should be adjusted to best highlight the colour of fibre you're attempting to isolate.
For more details, see the "image preprocessing" guide
To obtain an estimate of the count from pre-processed images:
$ source activate fluorescent-fibre-counting $ python -m count <directory of preprocessed images> results.csv
Running on the GPU
Since this project uses tensorflow, it's advisable to use the GPU if you've got access to an NVIDIA graphics card. This can speed up counting fibres by a factor of ~15 (e.g. from 1m 30 per slide to ~6 seconds).
To install with GPU support:
- find the installation instructions for your platform from https://www.tensorflow.org/install/
- follow instructions to prepare your platform for use with tensorflow-gpu
- initialise the environment with
conda env create --file environment.yaml