ktuite / FacialFeatures

Glasses, open mouse

FacialFeatures

1. run "make" to compile the codes
   

2. there are two modes for FaceFeatureDetect:

   Train new data
   

   ./FaceFeatureDetect -t <positive_urls> <negative_urls> <model_save_directory>

   Predict a well aligned and cropped image
   

   ./FaceFeatureDetect -p <hog_path> <model_directory>

Sampled data provided for training and testing:

 ./FaceFeatureDetect -t data/exp1_pos.hog data/exp1_neg.hog data/exp1
 ./FaceFeatureDetect -p data/pos.hog data/exp1 (outputs 1)
 ./FaceFeatureDetect -p data/neg.hog data/exp1 (outputs 0)

NOTE:
1. The train mode will not output anything.
2. The predict model will output a single number for the predicted label