What's the difference between bias/scale layer and eltwise layer when using add and mul function?

[ysy4...@gmail.com]

I found that bias/scale layer can perform addition and multiplication function, moreover, this two functions support broadcast operation while eltwise layer doesn't. So what's the difference between bias/scale layer and eltwise layer when performing add and mul functions (not broadcasting )?  Are they performing exactly the same thing?

[Przemek D]

Bias and Scale work by adding and multiplying a constant (or a vector or blob of constants) to their one input. This constant is a learnable parameter of the network.

Eltwise adds or multiplies two input blobs together - so an Eltwise layer with a PROD operation does almost the same thing Scale does, except the blobs it multiplies together are produced by other layers of the network, while in Scale one of the blobs is a learnable parameter.