josecanedo commented 4 years ago

Excellent paper and code! Thanks.

By analyzing the KCS matrix and reassembling the skeleton following the paper i got something like: 6 -> 7 -> 8 -> 9 7-> 10 -> 11 -> 12 7-> 13 -> 14 -> 15 which seems to be the upper body starting at the root (hip joint), but the lower bones assemble like this: 0 -> 6 0 -> 1 -> 2 3 -> 4 -> 5 I was expecting something like this: 6 -> 0 -> 1 -> 2 6 -> 3 -> 4 -> 5

Is there a error in the KCS matrix or I misinterpreted the skeleton?

KCS matrix

Ct = tf.constant([
      [1., 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1],
      [-1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
      [0, -1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
      [0, 0, 1, 0, 0, 0, 0, 0, 0, 0 , 0, 0, 0, 1, 0],
      [0, 0, -1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
      [0, 0, 0, -1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
      [0, 0, 0, 0, 1, 0, 0, 0, 0, 0 , 0, 0, 0, 0,-1],
      [0, 0, 0, 0, -1, 1, 0, 1, 0, 0, 1, 0, 0, 0, 0],
      [0, 0, 0, 0, 0, -1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
      [0, 0, 0, 0, 0, 0, -1, 0, 0, 0, 0, 0, 0, 0, 0],
      [0, 0, 0, 0, 0, 0, 0, -1, 1, 0, 0, 0, 0, 0, 0],
      [0, 0, 0, 0, 0, 0, 0, 0, -1, 1, 0, 0, 0, 0, 0],
      [0, 0, 0, 0, 0, 0, 0, 0, 0, -1, 0, 0, 0, 0, 0],
      [0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,-1, 1, 0, 0, 0],
      [0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  0,-1, 1, 0, 0],
      [0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  0, 0,-1, 0, 0]])

Aratrik commented 3 years ago

Strangely the last but one column has two ones, rather it should have a 1 and -1. I matched the matrix with standard MPII names I found it to be correct otherwise.

lvandoit commented 3 years ago

I noticed that in the discriminator these was a wassertein loss. But in the code it is defined as 'mean(y_true * y_pred)'. So there had any precess for it. @Aratrik @josecanedo

CHIKKIZZY commented 2 years ago

Strangely the last but one column has two ones, rather it should have a 1 and -1. I matched the matrix with standard MPII names I found it to be correct otherwise.

I count exactly one pair of 1 and -1 in each column. You may have counted along the row?

bastianwandt / RepNet

KCS matrix seems to be incorrect #7

KCS matrix