svhn's Introduction

SVHN

This repository contains 3 models for Street View House Number classification.
The first model is Histogram of Oriented Gradient combined with Neural Network which is contained in the folder HOG_ANN. The script train_hog_ann.py is used to train this model on training set and eval_hog_ann.py is used to evaluate this model on test set.
The second model is Convolutional Neural Network which is contained in the folder CNN. The script train_cnn.py is used to train this model on training set and eval_cnn.py is used to evaluate this model on test set.
The third model is Histogram of Oriented Gradient combined with Support Vector Machine. The script train_hog_svm.py is used to train this model on training set and eval_hog_svm.py is used to evaluate this model on test set.
To run these script first download the training and testing data from SVHN.
To train the CNN model run python train_cnn.py '--data_dir [path to data] --log_dir [path for log files]'
To train the HOG_ANN model run python train_hog_ann.py '--data_dir [path to data] --log_dir [path for log files]'
To evaluate the CNN model run python eval_cnn.py --model_dir [path to trained model] --data_dir [path to data]
To evaluate the HOG_ANN model run python eval_hog_ann.py --model_dir [path to trained model] --data_dir [path to data]
To train the HOG_SVM model run python train_hog_svm.py --data_dir [path_to_data]
To evaluate the HOG_SVM model run python eval_hog_svm.py --model_dir [path to trained model] --data_dir [path to data]

Recommend Projects