TAO pretrained object detection - detectnet_v2 error fixed

Hello everyone,
I was trying to follow the tutorial in TAO Toolkit Quick Start Guide, in which the a jupyter notebook was opened to train an object detection algorithm using detectnet_v2.
In section 2.D. Download pre-trained model :
The fourth cell "!ngc “https://api.ngc.nvidia.com…” is used to download the pretrained model from NGC, in case it doesn’t work you can go to the TAO pretrained object detection and copy wget cmd.
Anyway, this cell downloads the pretrained model “resnet_18.hdf5” into the folder : pretrained_resnet18.
The fifth cell “!ls -rlt $LOCAL_…” enters into a folder named “pretrained_detectnet_v2_vresnet18” which doesn’t exist.
To Fix this error, we should create a folder “pretrained_detectnet_v2_vresnet18” in the folder “pretrained_resnet18” and then download the pretrained model “resnet_18.hdf5” into the former one. Then change the model name to “resnet18.hdf5”.

This worked for me, I hope you can update the notebook so we can bypass this error.
Thank you!

Did you download the latest notebook?

tao-getting-started_v4.0.0/notebooks/tao_launcher_starter_kit/detectnet_v2

In 4th cell, it is as below. I can run it successfully.
$ ngc registry model download-version nvidia/tao/pretrained_detectnet_v2:resnet18 --dest $LOCAL_EXPERIMENT_DIR/pretrained_resnet18

1 Like

Thank you! I was working with an older version

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