Using German Traffic Sign Recognition Benchmark, in this script I have used Data Augmentation and then passed the images to miniVGGNet (VGG 7) a famous architecture of Convolutional Neural Network (CNN).
It contains 138 Million parameters and is a simplistic application of VGG-16 and VGG-19. In the end a model with an accuracy of 99.7 percent has been obtained.
ALl the requirements can be found in requirements.txt.
simply use pip install -r requirements.txt. You can also install tensorflow-gpu depending upon the
tensorflow version of yours, use any search engine for more information as Tensorflow-GPU
is a complex and large topic in itself.
The Jupyter Notebook contains all the relevant information about the Model.
The final model named model_final.h5 along with all the models trained can be found here.
A visual flowchart of model here:
I have used German Traffic Sign Recognition Benchmark Dataset or GTSRB, it can be downloaded from here.
The Jupyter Notebook should be referred for the Usage guidelines, while the Python script for a better understanding of documentation.
This Script has been prepared by Vybhav Chaturvedi