We've trained an LSTM that takes as input a sequence of length 10 of 22 body segment orientations represented with quaternions (22*4 = 88), consisting of a state vector of size 88. So overall my input is 10x88= 880 floats.
I'm having trouble figuring out how do initialize the Tensor class properly. I have all 880 floats stored in a float[] list called inputData.
If I try to initialize the tensor and later execute by:
Tensor output = m_Worker.Execute(input).PeekOutput();```
I get an assertion error in BarracudaCompute saying:
` Assert.AreEqual(X.flatWidth, W.flatHeight);`
with X.flatWidth = 88 and W.flatHeight = 10.
Can anyone offer me insight how to select the right tensor constructor arguments?
We've trained an LSTM that takes as input a sequence of length 10 of 22 body segment orientations represented with quaternions (22*4 = 88), consisting of a state vector of size 88. So overall my input is 10x88= 880 floats.
I'm having trouble figuring out how do initialize the Tensor class properly. I have all 880 floats stored in a float[] list called inputData.
If I try to initialize the tensor and later execute by: