Hi erik
Very nice! It works!
Question related to vec.dims: I run my model and got the following names and dimensions:
Inputs
Name: right, Type: DataType.FP16, Shape: [160, 120, 3, 1]
Name: left, Type: DataType.FP16, Shape: [160, 120, 3, 1]
Outputs
Name: output, Type: DataType.FP16, Shape: [160, 120, 2, 1]
It looks like the shape is [WHCN] and not [NCHW]. Is this because my model or the vec.dims uses different convention to report [NCHW]?
The reason I ask this question is because in my pipeline:
#only show left input to nn for brevity
manipLeft = pipeline.create(dai.node.ImageManip)
manipLeft.initialConfig.setResize(160,120)
manipLeft.initialConfig.setFrameType(dai.RawImgFrame.Type.RGB888p)
manipLeft.out.link(nn.inputs['left'])
Would manipLeft.out as [3,120,160] or [160,120,3]? and would this pipeline automatically add the extra dimension as [160,120,3,1] before feed in the nn? or I need to manually "unsqueeze" it? and how? And it seems the nn looking for FP16 as input, should I use:
manipLeft.initialConfig.setFrameType(dai.RawImgFrame.Type.RGBF16F16F16p)
instead? (I did try, but it failed to connect ; (
Please advise. Thanks a lot for your help again.