I am planning to use the Peoplenet model on an MCU. But the MCU supports only 1MB models and peoplenet has 8.32MB.
So I planned to prune the model further to reduce the size and then use tflite for micro to convert it to 1MB model. But the onnx file of Peoplenet model is unable to prune because DetecnetV2 pruning accepts only hdf5 models.
Please help in converting the tlt model to convert to hdf5 model
Please refer to tao_toolkit_recipes/tao_forum_faq/FAQ.md at main · NVIDIA-AI-IOT/tao_toolkit_recipes · GitHub.
I believe the provided code only converts etlt to onnx and not tlt to hdf5. Could you please help me in converting tlt to hdf5
It is the same. You can use it to convert tlt to hdf5 as well.
Can we convert etlt to hdf5? Could you help me with that
No, we cannot.
The .etlt file is actually encrypted .uff or .onnx file.
The .tlt file is actually encrypted .hdf5 file.
I need resnet34_peoplenet_int8 model’s hdf5 model to prune it, since detectnet_v2 supports only hdf5 model for pruning. Could you help me with that?
No, user can also run pruning with previous .tlt model.
Refer to DetectNet_v2 - NVIDIA Docs.
okay. How about etlt? Can I prune an etlt model?
We cannot prune an etlt model.
So basically for peoplenet model, you are not providing hdf5 or tlt of int8 model, not able to prune onnx model and not able to prune etlt model.
Please Suggest me a way ahead. I want to prune the resnet34_peoplenet_int8 model
In TAO, the resnet34_peoplenet_int8 onnx file is not for pruning.
You can prune the trainable model PeopleNet | NVIDIA NGC . After pruning, as mentioned in the user guide, retraining is needed.
There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.