Hello Friends,
I built a transfer learning Yamnet model below and saved as .h5 model. Then I used AI convert online to nb model, but I did not receive any result via email. I guess there is something wrong during convert my model. So could you please help having a look on my attach model and show me guide to what issue? or show me how to convert from a transfer learning Yamnet model.
my_yamnet.zip (1.9 MB)
Thank you in advance
Hi @Hai_Tran,
Thank you for sharing the model. It is possible that the issue is related to the model itself, which may have caused the conversion to be unsuccessful, and therefore no result was sent by email.
We are currently reviewing the model and conversion results and will update you once we have more information.
Thank you for waiting.
Hello Pammy,
I have been following this topic. Very appreciate your feedback.
Thanks.
Hi @Hai_Tran,
May I know how many classes your model was trained on? Our default setting is 512 classes. If the number of trained classes is fewer than 512, it may cause issues.
Hello Pammy,
Yes, I’m only using 8 classes. Has this convert tool ever used Yamnet Tensorflow transfer learning model? And share me any document to build and convert model?
If my transfer leaning model use 512 classes, then can it convert to nb?
Thanks.
Hi @Hai_Tran,
The conversion tool currently supports the YAMNet base architecture, which consists of 55 layers.
While reviewing the conversion process of your model, I observed the following error:
“ValueError: You are trying to load a weight file containing 1 layer into a model with 55 layers.”
It seems like an architecture mismatch, where the provided weight file does not correspond to the expected 55-layer YAMNet architecture, likely due to differences introduced by a transfer-learning model.
We do have documentation related to conversion and deployment only.