Built and trained a deep neural network to classify traffic signs, using PyTorch. The highlights of this solution would be data preprocessing, trained with heavily augmented data and using Spatial Transformer Network.
In this project, I will show you how to use PyTorch to classify traffic signs and how to imporve the classifier with a Spatial Transformer Networks. You will train a model so it can decode traffic signs from natural images by using the German Traffic Sign Dataset. After the model is trained, you will then test your model on new iamges of traffic signs from test dataset.
The model is designed using Spatial Transformer Network with a modified version of IDSIA networks and trained with heavily augmented data. The models reaches about 99.3% test set accuracy
My attempt to tackle this problem can be read in report.
Per the Anaconda docs:
Conda is an open source package management system and environment management system for installing multiple versions of software packages and their dependencies and switching easily between them. It works on Linux, OS X and Windows, and was created for Python programs but can package and distribute any software.
Using Anaconda consists of the following:
miniconda
on your computer, by selecting the latest Python version for your operating system. If you already have conda
or miniconda
installed, you should be able to skip this step and move on to step 2.conda
environment.* Each time you wish to work on any exercises, activate your conda
environment!
Download the latest version of miniconda
that matches your system.
Linux | Mac | Windows | |
---|---|---|---|
64-bit | 64-bit (bash installer) | 64-bit (bash installer) | 64-bit (exe installer) |
32-bit | 32-bit (bash installer) | 32-bit (exe installer) |
Install miniconda on your machine. Detailed instructions:
For Windows users, these following commands need to be executed from the Anaconda prompt as opposed to a Windows terminal window. For Mac, a normal terminal window will work.
git clone https://github.com/wolfapple/traffic-sign-recognition.git
cd traffic-sign-recognition
traffic-sign-recognition
. Running this command will create a new conda
environment that is provisioned with all libraries you need to be successful in this program.conda env create -f environment.yaml
source activate traffic-sign-recognition
conda env create -f environment.yaml
activate traffic-sign-recognition
At this point your command line should look something like: (traffic-sign-recognition) <User>:traffic-sign-recognition <user>$
. The (traffic-sign-recognition)
indicates that your environment has been activated, and you can proceed with further package installations.
conda info --envs
conda clean -tp
Now most of the libraries are available to you. Very occasionally, you will see a repository with an addition requirements file, which exists should you want to use TensorFlow and Keras, for example. In this case, you're encouraged to install another library to your existing environment, or create a new environment for a specific project.
To exit the environment when you have completed your work session, simply close the terminal window.
To uninstall the environment:
conda env remove -n traffic-sign-recognition
The GTSRB dataset (German Traffic Sign Recognition Benchmark) is provided by the Institut für Neuroinformatik group here. It was published for a competition held in 2011 (results). Images are spread across 43 different types of traffic signs and contain a total of 39,209 train examples and 12,630 test ones.
Download the dataset. This is a pickled dataset in which resized the images to 32x32.
Unzip the dataset into ./data
directory.
Run the script train.py to train the model.
python train.py
You can include one or more command line arguments.
python train.py --help
Various training options can be confirmed with the following command.
As the model trains, model checkpoints are saved to model.pt file to the current working directory. You can take one of the checkpoints and run:
python evaluate.py --data [data_dir] --model [model_file]
We want to visualize the output of the spatial transformers layer after the training, we visualize a batch of input images and the corresponding transformed batch using STN. Check the generated image file.