Is CHW format mandatory in inference ?

[aidamian]

So, the model was loaded/trained with Keras/Tensorflow and for sure input is HWC however during inference I see both in testing and conversion scrips we are forcing CHW input. Why is that and how is that possible?

[faustomilletari]

There is conversion going on. You have your tensor flow model in hwc and once it's in tensor rt everything will be chw. Fausto Milletarì Sent from my iPhone

…

On 7. Feb 2018, at 07:07, Andrei Ionut DAMIAN ***@***.***> wrote: So, the model was loaded/trained with Keras/Tensorflow and for sure input is HWC however during inference I see both in testing and conversion scrips we are forcing CHW input. Why is that and how is that possible? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[aidamian]

In understand. Does this mean when input tensor if 4D the convertor automatically assumes swapping of dim 2 and dim 4 (certainly for the case when input is tensorflow frozen graphs I mean) ?

Also, why here
https://devtalk.nvidia.com/default/topic/1025594/how-to-feed-a-3-channel-image-to-tensorrt/
they say you feed whatever you have ?

Thanks