Permalink
Cannot retrieve contributors at this time
Name already in use
A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
RoadsignDetection/README.md
Go to fileThis commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
29 lines (25 sloc)
1.44 KB
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# cnn | |
A convolutional Neural Network to detect german road signs in images of different sizes. | |
The networks layout is similar to the one described in | |
The neural network features the following methods: | |
* Usage of an advanced optimizer (ADAM algorithm for gradient descent) | |
* Convolutions to speed up training | |
## Installation | |
Conducting training of your own using the convolutional neural network from our repository requires different software: | |
* Python | |
* Ipython notebook (for preprocessing and postprocessing scripts) | |
* Tensorflow (for training the cNN on graphics cards) | |
* Matplotlib (for plotting) | |
Installation can be done easily with pip by | |
```bash | |
pip install tensorflow matplotlib ipython | |
``` | |
and ipython/jupyter notebook can be installed using the following guide http://jupyter.org/. | |
## Use the network | |
This repository includes several successfull training procedures which yield about 99% accuracy. | |
If you are interested in using them, have a look at the postprocessing script. | |
## Conduct training | |
If you want to conduct training yourself or change the existing code, you are welcome to do so. Preprocessing contains code to | |
read the images from the official GTSRB website. It also converts all images to a proper training and test set in which all images have | |
the same size, the same color layout (grayscale or YUV) and are rotated and scaled randomly to increase the training data size. | |
Several different training and test sets can be easily generated this way. |