cudnn without any framework

93 Views Asked by At

Does anyone have an idea, what is matched wrong. It compiles without any error, etc., but the Output goes always from Predict class 5 to 3 to 1 and doesnt change.

Code(important here PropagationHelper&Network&Layer):

Output:

Update Weight(Value of ConvGradConv is to huge!) enter image description here

Networkflow(first 5 and last 5 data in each layer) enter image description here

Okay one Error corrected(FullyBackward used inchannel insteed of inputsize), but there must be another one with the ConvLayer.

0

There are 0 best solutions below