In this Project, FDDB dataset is used and the script (Load_Dataset.py) to generate the dataset is inside the 'Data' directory.
Next, 'hshukla_Proj2_task1.py' - this script is run to display haar features. More details of the functionality of this script is mentioned below.
========================================== Task 1 === ========================================
To plot top 10 features of each of the 5 types of Haar-like feature and
display the error of all individual weak classifiers.
Step 1. Run "hshukla_Proj2_task1.py" to display errors corresponding to each of the 5 Haar-like features detected on the image and save the best features of all types.
Results of this code can be viewed in the "/Results/BestFeatures"
The snippet of the terminal after execution of this code is saved as
"/Results/output_task1.png"
Scripts for the following mentioned task reffered as Task-2:
1. python3 Train_ViolaJones.py --number_of_weak_clf 50
2. python3 Test_Viola_Jones.py --model 50
3. python3 FaceDetect.py : Contains the train and test functions
4. python3 hshukla_adaboost_viola_jones.py : Class written to perform
face classification using Adaboost technique.
========================================== Task 2 ========================================
# In this project, 6 models are developed and their corresponding pickle
files are saved in the "Models" directory. (2.pkl, 5.pkl, 15.pkl, 20.pkl, 50.pkl and 100.pkl)
# Users can directly run the "Test_Viola_Jones.py" file to evaluate the
performance of the trained model on Test Dataset. Users should run the
code from command line as follows:
Syntax -> python3 Test_Viola_Jones.py --model <Name of the trained model>
Example -> python3 Test_Viola_Jones.py --model 50
{Name of the trained model is 50.pkl, but extension is not required while giving command line input at the time of execution}
# DATASET :-> 100 positive and 100 negative images are used as the training images and are
located in "Data/Train/".
-> 20 posiitive and 20 negative images are used as Test images and are stored in
"Data/Test/"
-> Refer to "/Data/Load_Dataset.py" to generate the dataset required for
Training.
# TRAINING : Run "python3 Train_ViolaJones.py --number_of_weak_clf 50" from the command line
to train the Viola Jones classifier algorithm.
--number_of_weak_clf is the command line argument that specifies the number of
weak classifiers to train the model.
Resulting model will be saved in the project directory as "50.pkl" if the input
given by the user from command line is 50. The input must be an integer value.
As we increase the number of weak classifiers to train the model, the accuracy
keeps on increasing in the test dataset.
# HYPERPARAMETERS : Number of weak classifiers.
Test Accuracy on 25 Weak classifiers is 85%, whereas it is 92.5% when number
of weak classifiers is 50