I have trained a re-id model using tao . now i wanted to convert the pth model generated to etlt format to run in deepstream . How can i export ?
Please refer to TAO doc or TAO notebook( tao_tutorials/notebooks/tao_launcher_starter_kit/re_identification_net/reidentificationnet_resnet.ipynb at main · NVIDIA/tao_tutorials · GitHub ). The result will be an onnx file.
The .etlt model is actually an encrypted version of onnx file. It is not needed to export to etlt file now. You can directly export to an onnx file and then set an onnx file in the deepstream config file.
where can i get the tlt reidentification model ? In the NGC catalog i could only find onnx and etlt file from deepstream.
You can find the .tlt format pretrained model in ReIdentificationNet | NVIDIA NGC.
But please note that it is used to run training. It is not a deployable model.
i needed it for finetuning purpose.
The .tlt format can be used to finetune.
when i have used tlt model i am getting PermissionError: Cannot access model state dict without the encryption key. where should i add the encription_key and what is it? adding encryption_key: “tlt_encode” in the spec file gave me error : Error merging ‘spec_file.yaml’ with schema
Key ‘encryption_key’ not in ‘ReIDModelConfig’
full_key: model.encryption_key
reference_type=ReIDModelConfig
object_type=ReIDModelConfig
the link is tao version 5.x but i am currently using 6+ and it doesnt support encryption_key in spec file and i am facing issue for encryption.
In TAO6 spec files, https://github.com/NVIDIA/tao_tutorials/blob/main/notebooks/tao_launcher_starter_kit/re_identification_net/specs/experiment_market1501_resnet.yaml#L2, there is also encryption_key. Could you double check your yaml?
Thank you for the reference . the issue was using encryption_key: nvidia_tao , inside the model parameter.