pixel representation

[gravitychen]

Dear author,

the code in QuaternionLinear() splits 1d input data into R,I,J,K part , but when we want to do some task on images e.x.MNIST dataset size = ([1,784]), it's weird to operate pixels like this, if I want to make each pixel as quaternion , how should I change the code?

Thank you
Chen

[TParcollet]

MNIST is on the grayscale. Thus, you can replicate each value three time to obtain 0 + value I + value j + value k.

[gravitychen]

MNIST is on the grayscale. Thus, you can replicate each value three time to obtain 0 + value I + value j + value k.

Thank you ,so it's better to use QuaternionConv() right?