docs | downloads (stat) | faq | trac | wiki | dev. mailing list | svn repo. | awards | screenshots

Frequently Asked Questions (FAQ):

How to retrain the Kepenekci Face Recognition module? (asked by Livio Lima on Jul 13, 2009)

Open rtftr/src/RtftrUserInterface.cc . In Line no. 804 (find this comment "//---------- Initialize the Kepenekci Face recognition module ----------"). Follow the steps given below:

1. store your training images (64x64 size, grayscale) in rtftr/data/kepenekci/trainfaces/. The naming of training images is done as:

// f0001v001.pgm - face-id = 0001, pose-id = 001

2. In line no 807, change the filename of training data to whatever you wish

string trainDataFilename = "Kepenekci_TrainData_July14_2009.dat";

3. Now as you have new training images, you have to invoke the Kepenekci.PerformTraining(...) method in Kepenekci::RETRAIN_OVERWRITE mode (creates a new data file if specified data file does not exist). There is already a commented block in line no. 813. Uncomment it and then comment the code in line 814.

kepenekci->PerformTraining(Kepenekci::RETRAIN_OVERWRITE, trainWindowSize, trainImgDir, trainDataDir, trainDataFilename);

// kepenekci->PerformTraining(Kepenekci::NO_RETRAIN_RETRIVE, trainWindowSize, trainImgDir, trainDataDir, trainDataFilename);

4. Execute the program, load a video file and press "Start" button. It will take some time to retrain the Kepenekci database. Then the recognition process starts with new database.
NOTE: You do not need to retrain everytime. You can use the data file (created after you execute the code for the first time). So to reuse the data file now invoke the Kepenekci.PerformTraining(...) method in Kepenekci::NO_RETRAIN_RETRIVE mode as shown below

// kepenekci->PerformTraining(Kepenekci::RETRAIN_OVERWRITE, trainWindowSize, trainImgDir, trainDataDir, trainDataFilename);

kepenekci->PerformTraining(Kepenekci::NO_RETRAIN_RETRIVE, trainWindowSize, trainImgDir, trainDataDir, trainDataFilename);

We are planning to embed the retraining task of both Kepenekci and subspace LDA modules in our GUI and it should be the part of our next stable release(rtftr-0.2). Meanwhile, you can achieve this using the steps described above