The problem of feature dimension

[LuMenCat]

In ./datasets/CASIA_CK+_data_gen.py file:
you set the dimension of sample_numpy is 4, then through some concat operations, I did not see where the dimension could be changed. But in data_normalization(), the dimension of Data is 5.

I feel a little confused here. Can you answer it？
Thanks.

[negarhdr]

sample_numpy represents the landmarks of each video/image_sequence and its 4 dimensions represent the number of frames, number of faces in each frame, number of landmarks in each face, and the feature dimension (channels) for each landmark!
After the concatenation of several samples in a fold, we will have an extra dimension "N" which shows the number of concatenated samples! So the total number of dimensions will be 5.

[LuMenCat]

This is the mistake when I run this code.
On line 38 of CASIA_CK+_data_gen.py, you create a 4-dimensional array, then through the following concat operations,such as line 49, 97. But np.concatenate() could't increased the dimension. And in data_normalization(), the dimension of Data is 5, can you introduce it in detail？And describe what N, V, C, T, M represent respectively.

Thanks.

[negarhdr]

You're right. Thanks for raising the issue. It should be fixed now.

The dimensions are as follows:
N: number of samples,
C: number of channels (dimensions) of each facial landmark
T: number of frames in an image sequence
V: number of landmarks in each face
M: number of faces in each frame