I’m new with image classification, recently i try to try a tutorial from tensorflow, Классификация изображений с помощью TensorFlow Lite Model Maker
i have a question about optimizer, can i change the optimizer? How can i use another optimizer based on the code from image?
Reading the API I do not see a way to change the optimizer. You do not achieve good accuracy?
No, i just need it for comparation, i have a task from my college about image classification and i asked to do it with different optimizer to evaluate which one is better.
So I presume that you do not need it for mobile or IoT device. For now, you have to do it without using model maker then
I also have to do it with android, that’s why i need to train the model using mobilenet
For this version of Model Maker I do not see an option to change the Optimizer. So do it the old way and convert the Saved Model with TF Lite converter.
here is a good tutorial to help you: Image Classification with TensorFlow Hub
Hi @jeri_shanks , currently TensorFlow Lite Model Maker doesn’t provide such customization.
We are only targeting at easy-to-use interface for creating a model, instead of swapping every bit of modules. Usually, changing learning_rate and epochs will be the option to tune model quality.
(BTW, if you are interested in adding this feature, you can contribute code to Model Maker’s repo, and make it work in nightly version.)