Jetson Nano (card-image-r32.4.2) and Tensorflow 1.13

[filmo]

Following the nano setup instructions no longer seems to work with current links. In particular, the current (a/o 2020-05-25) SD card nv-jetson-nano-sd-card-image-r32.4.2 comes with Cuda 10.2 on it which is not compatible with the in the subsequent install of TF in the instructions.

pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v42 tensorflow-gpu==1.13.1+nv19.3

Question: what is the correct archival SD card that is compatible with TF 1.13 ?

Alternatively, which version of TF is compatible with the current Cuda 10.2 and also works with the Donkey Car code base? (For example, could we use the TF 2.x family ? How about TF 1.15 instead of 1.13 ??)

[tikurahul]

Nvidia updates it images every couple of months. https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/index.html is the best place for information about how to install Tensorflow on Jetson.

Specifically the step you are looking for is:

pip install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v44 ‘tensorflow<2’

This should install TF 1.15.2 on the Nano. You should also refer to the above URL for the pre-requisites.

[tawnkramer]

@tikurahul do you want to change the docs?

[tikurahul]

I also need to fix keras.py. Will send you a PR today.

[NathKorat]

you saved my life!!!!

[Ezward]

Closing as obsolete.