pvlbzn / kot-or-not

Logistic regression (shallow neural network) which predicts whether input image is a cat or not.

Kot or Not

Kot or Not (from Russian word Кот, a Cat, wordplay) project is an implementation of a logistic regression algorithm using numpy for matrix arithmetics.

Algorithm uses sigmoid function, forward and backward propagations as well as vectorization techniques.

Data

Current implementation has training set with cardinality 209. Images scaled down to 64 x 64 x 3 and processed as a feature vector (64 * 64 * 3, 1).

Usage

usage: main.py [-h] [-d] [-r] [-i INPUT]

Cat or not neural network

optional arguments:
  -h, --help            show this help message and exit
  -d, --dump            use model dump
  -r, --retrain         retrain model
  -i INPUT, --input INPUT
                        path to input image

To classify your image first you need to train a model

python3 main.py -r

The model will be dumped into a flat text files: weights column vector and bias literal.

After training --dump can be used together with --input image path

python3 main.py -d -i images/moon.jpg

Baby can Moon is certainly a cat, and kot-or-not NN things the same.

Roadmap

• Implement a self learning feature from input images