Keras vgg16 model to try/engine

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

his @Morganh

I am having a trained Classification model VGG16 … which I have trained in keras. its extension is .h5.
How can I convert it into .trt or .engine file so that I can use it on Jetson Nano for running predictions.
Any help will be appreciable?
thanks

2

Please run TLT classification network. Set it as the pretrained model.
Then get tlt model and etlt model, and use tlt-converter to generate trt engine.

1 Like
3

ok thanks I can try this .

4

@Morganh
it throws error
NotImplementedError: Invalid model file extension. /workspace/tlt-experiments/classification/pretrained_resnet18/tlt_pretrained_classification_vvgg16/vgg_16.h5
do I need to rename the extension as .hdf5 from vgg_16.h5.
is it the right way ?

thanks

5

Officially TLT provides the pretrained model in Transfer Learning Toolkit — Transfer Learning Toolkit 3.0 documentation

For your case, you can can rename it to .hdf5 and retry.

1 Like
6

@Morganh
Any clue about this .

Screenshot 2021-03-31 at 2.18.23 PM
Screenshot 2021-03-31 at 2.18.23 PM2104×754 194 KB

at the time of training in keras
keras version is 2.4.3
and tensorflow version is 2.4.1

tried but did not get any solution for this .
thanks

7

So, your pretrained model is not compatible with TLT.
Please use the pretrained models in ngc.

1 Like
8

got it thanks .