Academic reference

Essay of Project

Environment

we worked over TensorFlow 1.4 and anaconda environment. a .yml file with dependencies will be uploaded soon.

Hands on

This bullets are consecutive steps we suggest to perform for easier diving-in:

First, we are available for any question and support here and on [email protected] - so don't be shy.
focus on cucu_train.py only. read the code that deals with creating the model and relevant data. it is self explanatory.
we bring to notice that many sections are stiil to-be-extracted to methods etc.
Run it and handle all env. obstacles which will get to you:)
now, once you are ready to play with all the parameters of our project open cucu_config.py - you got there anything you need to control NN hyper-parameters and data-generating parameters.
- we suggest to nevigate to original config file of Mask RCNN where hyperparameters definitions are more elaborated.
Next we suggest to explore our project_assets folder.
- There, you'll get aqcuainted with our classes for generated dataset creation, real, and hybrid (cucu_classes.py).
- In cucu_utils.py we poured core-functions for generating synthetic images of crops.
Now, you should be ready for playground.py where you can run different metrics on a small test-set to benchmark your trained NN.
Lot's of analysing functionalities where imported and upgraded from here