README.md

DeepLearning-ComputerVision-GermanTrafficSigns

Pushing German Traffic Signs Recognition Benchmark (GTSRB) beyond human performance level(98.84%)

Background: This repo builds on Udacity project 2 and attempts to republicate end to end machine learning without manually designing extensive features extraction features increasing efficiency

The motivation of using Deep Learning for GTSRB to improve accuracy is in (Traffic Sign Classification Using Deep Inception Based Convolution Network) Mrinal Haloi's paper which includes more than 6 convnets

Goal 1: To learn and build deep learning model(e.g. Inception, VGG, Resnet) to improve GTSRB's performance

Goal 2: Improve interactive visualisation above and beyond Udacity project 2(widgets) leveraging Bokeh

Goal 3: Provide a machine learning platform to integrate reinforcement learning into supervised learning network