Tmn(q) v3,v4,v5,v6,v7,v8

[ThorKing01]

@sergiomsilva About the reconstruct function of keras_utils.py in the code,affine=[ v3,v4,v5,v6,v7,v8],A=np.reshape(affine,(2,3)),A[0,0] = max(A[0,0],0.)A[1,1] = max(A[1,1],0.). It's means A[1,1]=v7, if v7>0. But v6 should compare with 0 in the Loss Function in the paper, so I hope you can help me understand the difference. Thank you very much!

[tailtq]

@ThorKing01 I figured out that his implementation follow this style: pred0, pred1, pred2 should be understood as v3, v4, v7 pred3, pred4, pred5 should be understood as v5, v6, v8

Then he multiplied the tensors by [-v, -v, 1., v, -v, 1., v, v, 1., -v, v, 1.].

1. is for plural.