Details here.
This challenge requires us to identify the pixels that belong to a "cell" in rectangular masks that vary in dimensions. Each mask contains dozens of cells, and our script must be able to identify all instances of cells and their corresponding pixels.
$ git clone https://github.com/tylerhslee/kaggle2018
$ cd kaggle2018
$ pip3 install -r requirements.txt
$ chmod +x main.py
$ ./main.py
Before running the script, make sure to point the config.yml
file to the correct dataset.
I plan to take the following steps to approach this problem.
-
Isolate each instance of cell image from every mask and save them as separate .png files.
-
Train Mask R-CNN model to implement Tensorflow's object segmentation algorithm. (Note to self: Tutorial to follow)
-
Run the instance segmentation model on the masks originally given to us.
Currently, the project has completed step #1, and is now preparing to work on #2. I'll shortly update the script to generate correct labels for each mask that can be used for training the model.