rort1989
commented
7 years ago There are 229 sequences for training in the format of array of cells. Each cell contains a training sequence of 177xT matrix. 177 is the dimension of features (landmark points, Gabor wavelet coefficients and LBP + PCA) extracted from raw image. The details are in the paper mentioned in the Readme. T is the number of frames in the sequence.
For testing data, it is a 177x304 matrix. 177 is a result of the same feature extraction process. 304 is the number of testing frames. Notice that the testing is performed on individual frame rather than the whole sequence.
In general, you can use your own feature extraction pipeline to form N-dimensional feature and apply the method.
In the example you give, you have 299 sequences for training, right? Each training sequence is saved as a 177x14 matrix. How does the dimension come from? I would assume 14 is the number of frames. Then 177 is the columnized frame, right? Or just the feature dimension?
Similar, the test sequence you give is saved as a 177 x 304 matrix. Once again, how do you process a frame as a 177 feature vector?