zepheus/ml-traffic

Traffic sign recognition framework used for Kaggle competition

Trafic Sign Recognition

This project will execute the training and prediction of traffic signs, based on the Kaggle competition Here.

It uses both a convolutional neural network (Lasagne) and Logistic Regression (scikit-learn) in combination with feature extraction through scikit-image. The models were trained on an NVidia GTX 960 with 2GB of memory.

Executable files:

The executable files are listed below with their respective functionality:

• haar_importances.py:
  • This file will calculate the importances of the different haar configurations.
  • All importances will be written to the file "haarImportance.txt" in the current directory.
  • The importances are sorted according their importance in descending order.
  • This means that the most important configuration will be on the first line of the file.
  • This file can later be used by the haar_feature.
• main.py:
  • This file will train the model on all the train given and predict the results of the test images given.
  • These results are all written to a file named 'result.csv' in the current directory.
  • The given train and test images are specified in the python file. Also the used model and features are given in the code file.
• meta_parameter_estimators.py
  • This file will test metaparameters of features based on the error_rate.
  • Which parameters, features are trained with which trainer is specified in the file itself.