Great tutorial, I'd add that GraphModel is not fine-tunable, so you can not re-train it. The pb models that convert to GraphModels are then only useful for inference (to test the model). I think it is said but just to make it obvious, if you want to fine tune / re-train / use transfer learning, you need Layer Models and for that you need "tf-models-here.h5" or the js versions.
Yes I totally missunderstood what he meant with "transfer learning ability" I guess that was that layer model => transfers learning ability => fine tunable graph model => execution speed => not tunable
Although I've been enjoying this series, it is disheartening that this particular video that was made by Google developers and uses Google Colab no longer works after only 3 months. When following the instructions on Colab, the code returns a "TF-TRT Warning: Could not find TensorRT" error when attempting to run the tensorflowjs_converter. I can't find any resolutions to this issue, and can't move forward if we're going to be building on these concepts in the next video.
I have the same issue. I also couldn't run it on the WSL2 setup that had everything installed. Looks like even with the errors it creates the files. I was able to continue the series.
Catch more episodes from Machine Learning for Web Developers (Web ML) → goo.gle/learn-WebML
You are perfect! For a long time I tried to run it locally on my windows system but failed every time. You saved my with google collab. Respect!
Glad it was useful for you!
Great tutorial, I'd add that GraphModel is not fine-tunable, so you can not re-train it. The pb models that convert to GraphModels are then only useful for inference (to test the model).
I think it is said but just to make it obvious, if you want to fine tune / re-train / use transfer learning, you need Layer Models and for that you need "tf-models-here.h5" or the js versions.
Yes I totally missunderstood what he meant with "transfer learning ability" I guess that was that
layer model => transfers learning ability => fine tunable
graph model => execution speed => not tunable
Although I've been enjoying this series, it is disheartening that this particular video that was made by Google developers and uses Google Colab no longer works after only 3 months. When following the instructions on Colab, the code returns a "TF-TRT Warning: Could not find TensorRT" error when attempting to run the tensorflowjs_converter. I can't find any resolutions to this issue, and can't move forward if we're going to be building on these concepts in the next video.
I have the same issue. I also couldn't run it on the WSL2 setup that had everything installed. Looks like even with the errors it creates the files. I was able to continue the series.
same problem, did you find any solution please ?
@@lokidarkbeard2647 same problem, did you find any solution please ?
@karimkalboussi4703 even though it gives errors, it still produces the files you need.
I get the following error: Exception: Error dumping weights, duplicate weight name kernel