Closed gcmike closed 4 years ago
@gcmike I have the same issue on the zero, could you find a solution?
Same here, looking for a solution
@marsmith I could solve my issue with my LeNet-5 by turning off biasing in Conv2d operation.
See here: https://github.com/tensorflow/tensorflow/issues/24732
Hi there. I have successfully install v1.13.1 on pi zero with the armv6l.whl. I also have keras installed and ran an MNIST MLP example correctly with KerasMNIST repo from EN10. But when I tried the cnn version of MNIST, some error regarding _FusedConv2D popped up. A bit researching suggests that the error is with the architecture. I have cross-checked with pi 3 and verified that the repo works on armv7. According to the comment in the second link, cross-compiling with some setup will make it work. I was wondering if it is possible to release a new one with this feature?