I am trying to cross compile tensorflow on host for nvidia tx2 now, I followed the instructions provided in the bellow link but i am not able to compile successfully.
Does anyone success cross compilation of tensorflow on host? can you give me some help?
Could you tell us why you want to cross-compile Tensorflow on the host?
Here are some TensorFlow installation tutorial (device) for your reference.
Also some public wheel for you:
Thanks for your help and the reason i want to cross-compile Tensorflow on the host is maybe we will modify Tensorflow codes, and every time compile it on tx2 maybe slower than host.
Thanks for your feedback.
We don’t have experience on cross-compiling the Tensorflow libraries.
Maybe other users can share their experience.
Sorry for the inconvenience.
Ok, Now i can cross-compiling the Tensorflow libraries according :
i would like to compile application that based on tensorflow and opencv, do you have and idea of how to do that, and regarding the link you mentioned above can you tell me which compiler file you choose from the list ?
compile.sh Clones and build tensorflow for current machine. Takes the tensorflow version as a parameter
cross-compile.sh Generic script for cloning and building tensorflow using a cross toolchain. Launch it without args for usage
compile-arm.sh Clones the raspberry/tools repository an launches a cross compilation using the toolchain in it. Takes the tensorflow version as a parameter
compile-android.sh Download Android NDK and launch a crosscompilation for android using it. Takes the tensorflow version, android arch and package version as parameters
compile-ios.sh Clone and build tensorflow for iOS with the C API.
create-pkgconfig.sh Generic script for creating a pc. Launch it without args for usage
Thanks in advance
You don’t need to cross-compile you application. Just install the required package.
Here are some information for TensorFlow and OpenCV package on Jetson:
I think the TF pip package doesn’t all libraries for a native TF-application, especially those (depended)third-part libraries.