dl-shapes

Using DL to solve shape classification and shapes counting problem.

Task

1. Shape classification: network outputs six numbers: probabilities of a class being found on the image (in at least one copy).
2. Geometric shapes counting: network outputs 10 probabilities for each class, representing different numbers of objects of this class on the image. So the network should have 60 outputs. Outputs from 0 to 9 should sum up to 100%, so outputs from 10 to 19, and so on.

Files

• classification.ipynb - Jupyter notebook with the report related to the first part
• counter.ipynb - Jupyter notebook with the report related to the second part
• main.py - python script for running the experiments
• models - directory with ML models for both tasks
• datasets - directory with python scripts needed for various operations on the dataset

How to run

You'll need pipenv.

pipenv install --dev
pipenv shell

Now you can run the main script.

python main.py -m classifier

You can see all arguments here:

usage: main.py [-h] [-n] -m MODEL [-e] [-f FILE]

Work with shapes networks.

optional arguments:
  -h, --help            show this help message and exit
  -n, --neptune         use neptune.ai for logging
  -m MODEL, --model MODEL
                        which model should be trained
  -e, --entropy         development or training environment
  -f FILE, --file FILE  file for storing output for plotting