An artistic project written in JavaScript & Processing 3.5.4
A collaboration between KDArslan and GiorginoSerbuciano.
Licensed under the GNU General Public License v3. Copyright 2021.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
We do not collect any information from you. For privacy concerns, please consult the websites of the libraries used by this software:
WE TAKE NO RESPONSIBILITY FOR ANY LOSS OF DATA.
'October' is an art project that makes use of machine learning. It consists in a simple interface (p5.js & html component) and a neural network for image classification (ml5.js component) trained on an algorithmically-generated dataset (Processing component). The interface allows user input in the form of a mouse drawing. This is passed to the neural network, which classifies the user's input as belonging to one of two categories. The final output, altered by the neural network's classification, is a response to the user's input.
In order to run this web application, you will need:
- An up-to-date web-browser;
- An internet connection;
- A running local server, see https://github.com/processing/p5.js/wiki/Local-server#vs-code-live-server;
The interactive portion of the software corresponds with sketch.html. The scripts loaded in this file are contained in sketch.js and controls.js. Running this file on your local server will load the interface and a pre-trained NN model. Making full use of this portion requires no knowledge of p5.js.
For creating your own data set using our code, you will need Processing (3.5.4 or greater) installed on your machine. Once installed, you can run dataset.pde to generate your data set. Make sure to format your file names as follows: label_001.png. label should belong to the categories that you are training the NN for; this will be read when loading for training. This process presupposes familiarity with Processing, p5.js and ml5.js.
We have provided you with two versions of our model. To create your own model, run train.html on your local server. The corresponding script is training.js. If you have placed images inside the directory dataset/data, these will be loaded automatically. To run the training, either press "t" on your keyboard or type startTraining() in the console. Please keep in mind that training a model with large amounts of high-resolution images may be very resource-intensive and/or take a long time.
The interface consists in a black screen (the drawing surface) and a few overlay elements (the drawing area, the instructions, the title). The drawing surface is used for both input and output. The output is presented after and on the basis of the input. This output may be stored to disk in the form of an image file.
Corresponding files: sketch.js, sketch.html, controls.js
The machine-learning component consists of a neural network configured for the classification of images size [70, 70, 4] (4 color channels). This is an ml5.neuralNetwork() object. Training is currently configured to run for 50 epochs with a set size of 200. The model is trained in train.html. Once training is over, the resulting model files are saved to disk. If these files are placed in the folder model/, they will automatically be loaded in sketch.html.
Corresponding files: training.js, train.html, all of directory model/
The data component consists of a Processing script. This script generates a set number of drawings and saves them to disk every frame. Running this script will write files to disk under dataset/data/. Please beware that this process may be resource-intensive and/or take a long time and that no over-write protection is implemented.
Corresponding files: all of directory dataset/
.
├── controls.js
├── dataset
│ ├── data
│ └── dataset.pde
├── LICENSE
├── model
│ ├── entire_bias
│ │ ├── model.json
│ │ ├── model_meta.json
│ │ └── model.weights.bin
│ ├── lobed_bias
│ │ ├── model.json
│ │ ├── model_meta.json
│ │ └── model.weights.bin
│ ├── model.json
│ ├── model_meta.json
│ └── model.weights.bin
├── README.md
├── sketch.html
├── sketch.js
├── train.html
└── training.js