/osr

Optical Sketch Recognition in TensorFlow

Primary LanguagePython

Optical Sketch Recognition

OSR uses a deep convoluted neural network implemented in tensorflow to identify sketches of graphs

Overview

  • ogr.py runs a desktop sketching application that classifys the graph

  • ogr_web.py starts a webserver that can be sent an encoded sketch in a POST request and returns the classification

  • ogr_trainer.py trains the CNN and saves it to models

  • create_corpus.py is a program that produces a set of ~500 labelled examples to train the CNN on

  • ogr_cam.py takes a picture of a graph from the webcam and feeds it into the recognition software. It's temperamental!

Prerequisites

To run the trainer you require tensorflow, Tkinter and numpy.

To run the image recognition you also require opencv.