Tao classification inference errors

Please provide the following information when requesting support.

• Hardware A6000
• Network Type Classification
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here)
• Training spec file(If have, please share here)
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)
I am trying to run this file tao_tutorials/notebooks/tao_launcher_starter_kit/classification_tf1/tao_voc/classification.ipynb at main · NVIDIA/tao_tutorials · GitHub with mobilenet_v2

I have performed every step up to the model inference cell. I keep getting an error saying that a results.json is not there. I looked in datasets and there is no file. What to do?

!tao model classification_tf1 inference -e $SPECS_DIR/classification_retrain_spec.cfg
-m $USER_EXPERIMENT_DIR/output_retrain/weights/mobilenet_v2_$EPOCH.hdf5
-b 32 -d $DATA_DOWNLOAD_DIR/split/test/person
-cm $USER_EXPERIMENT_DIR/output_retrain/classmap.json

I get this:
FileNotFoundError: [Errno 2] No such file or directory: ‘workspace/tao-experiments/data/split/test/person/result.csv’

May I know if the datasets are available before training?

How about the training? Is it successful? You can check the training log and also check the result folder.

Is it missing / in the beginning? /workspace?

thank you for the advice. I was able to troubleshoot and have the cell run. I have completed that notebook and created to ONNX file.

I have 2 more questions if you can help?

  1. can I use mobilenet_v2 in the pytorch classification container?

  2. if not, then can I do data transforms with the tao classification_tf1 container? I need to do flip and rotate on images

  3. this notebook did not create a .tlt file. Do I need that model format?

For pytorch docker, the available backbones are mentioned in Image Classification PyT - NVIDIA Docs. The mobilenet_v2 is not supported.

There is a flag disable_horizontal_flipmentioned in Image Classification (TF1) - NVIDIA Docs. You can do filp augmentations. For rotation, it is not supported by default.

Since TAO5.0, it will generate .hdf5 instead. Previous .tlt is actually a decrypted file.

Thanks for the details! You saved me a lot of time reading documentation. I am thankful for you quick responses.