Test watershed technique

[TerenceTYAranowitz]

Description

Test a rough implementation of watershed

Files

augmentation.py data_utils.py

Tasks

-[ ] make sure data_utils class returns proper time of image -[ ] create rough implementation of watershed -[ ] createe new augmentations as needed

[JarrodRShepherd]

Description

We have found a consistent way to find a dynamic threshold value for the images. Now, we have to fine tune the parameters of the dilation and erosion steps before the watershed method is called. The images below are examples of bad watershed results. The first pair detects two extra small focci points. The second pair detects one large focci point when it should be separated in two.

Images

Tasks

• [x] Experiment with different parameter values for dilation and erosion
• [x] Determine the parameters that produce the best results

[nadia-eecs]

I would actually put watershed in a separate file (eg. watershed.py) in the models directory because it isn't really an augmentation. In fact it is a strong baseline that you can use for unsupervised semantic segmentation approaches.