Read this document to understand the project layout.
Author: Rohini Sharma - [email protected]
Given an image of a mathematical expression, this program will identify the symbols in the image and evaluate the expression. The symbols are identified based on a default pretrained K-Nearest Neighbors model, or other model specified by the user. The ExprTk library is then used to evaluate the expression and return its value. The default model was trained on the CROHME dataset. It can detect the following symbols with 85% accuracy:
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, +, =, -, (, ), [, ]
Install Cinder at ~/Cinder, using the above link. Create a directory called my-projects in ~/Cinder. Clone this repo into my-projects. This project was developed in CLion on MacOS, and is not necessarily optimized for Windows.
To run this program, run the cinder-myapp executable located in the
cmake-build-debug/Debug/cinder-myapp/cinder-myapp.app/Contents/MacOS/ directory.
In order to train a new KNN model, run the executable with the --train flag. Use --training_images to specify
the directory for training images. Use --training_labels to specify the training label text file. Each image in the
training directory must be in the format X.JPG where X is the line number of the text file that has the
corresponding label for the image. Use the --model_save_path flag to specify the path where you would like your new model
saved.
If --train is passed without training images or training labels, assets/training/ is the default path for
training images and assets/train_labels.txt is the path for training labels. If
--model_save_path is not specified, the model will be saved at assets/my_model.xml
To evaluate an equation, pass the path to the equation using the --equation flag. If no equation is passed, a default
image will be used. If you wish to use a model other than the default KNN Model, pass it to the program using the
--model flag. This model must be an .xml or .yml format.
Train a model using default training data path:
cinder-myapp --train
Train a new model with images in ~/my-training-images/ and labels at ~/labels.txt:
cinder-myapp --train --training_images ~/my-training-images/ --training_labels ~/labels
Evaluate default expression image:
cinder-myapp
Evaluate expression image at ~/my_equation.jpg using model at ~/my_new_model.xml
cinder-myapp --equation ~/my_equation.jpg --model ~/my_new_model.xml