I want to convert python model to pytorch lite, but I got above error.
TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can’t record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
My code is simple,
class CNN_forward(torch.nn.Module): def __init__(self): super(CNN_forward, self).__init__() self.cnn= nn.Sequential(*list(models.mobilenet_v3_small(pretrained=True).children())) def forward(self, pixel): raw_data = pixel.transpose(1, 2) pixel= torch.squeeze(pixel, 0) preproc_data = self.batch(pixel) return preproc_data
I have to change axis of 1 and 2 because my java code generate tensor as
[1 x RGB x Num frames x H x W] so it can not pass CNN
so i changed to [1 x Num Frames x RGB x H x W]
In my case Num Frames is only 1
But They say some warnning.
Shold I change my python model?
Also, After run on mobile, I got this error, as same as below
What is this error comes from?
There is no batch size if run on mobile?
OR when num frames is one, there dimemsion is deleted on mobile?