Why Nvidia Tao Toolkit documentation is very bad?

I am trying to train a object detection model with TAO and deploy it to Deepstream on Jetson devices. However, the documentation is so bad. Let me explain why I think like that.

When I click to official documentation link of TAO, firstly I tried to train a model with google cloud. I sometimes can do that but every month it changes. It sometimes work sometimes don’t work.

The colab links in that page is not working now for example, however last month it was working and I trained a model. Now its not working because of the update in Google Colab again I think. There is no information or quick announcement that inform developers whats going on, and we spend days to figure out whats the problem.

Then I decided to move on with Google Cloud or other Cloud training techniques because I am tired of colab. Now I need a guide how to do it but in documentation nvidia don’t explain any processes. Worse than that, In following link lots of links are broken :)

I click to Brev quick deploy links but none of them are working. Broken links on official documentation of billion dollar company.

Also I tried to download a container to use it on Google Cloud Platform, but quess what? Links on documentation are not working of course.

Like this one :

nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf2.11.0

So, I am working on a project that may use many nvidia jetson devices, however I am afraid about the support to TAO toolkit or Deepstream will end. Is there any possibility that Nvidia stop maintaining these projects?

1 Like

Previously the google colab can work. It is a regression issue recently.

This is a docker name. It is not a link. You can find it under TAO Toolkit | NVIDIA NGC as well.
You can docker pull it.

We do not stop maintaining these projects.

Could you please run training on local machine or GCP VM machine?

I can open the “Brev quick deploy link”. Could you please double check on your side?

For colab, please rerun. We fix it by doing some changes on two files. It is already merged. You can run the colab as previously.
The detailed info can be found in [BUGFIX] Fix issues caused by pip version and mpi4py version · NVIDIA-AI-IOT/nvidia-tao@5c9a6b0 · GitHub.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.