Continue training from checkpoint using the model maker APIs

Simply put, I plan on training an object detection model using the simple model maker APIs, and after having trained on a batch of images, I exported my model. Now, I feel as if the model needs some more fine-tuning, and may as well require a lot more iterations like this in the future, not very far apart in the temporal space. Hence, I need to be able to retain the training checkpoints from the first batches so I do not have to train the images that it’s already trained on, over and over again till eternity. This would usually not be a problem, but since I am building an object detection model, and I found that a little demanding to code, I turned to the model maker APIs which seem to make life drastically easier. The problem now, however, is that I don’t seem to have a way of “saving my progress” throughout the training cycles.

Now, there’s the ‘spec’ parameter of the create method that might point in the direction of a potential solution, but I can’t seem to find anything. I tried uploading my model to a server and passing it’s URI as the ‘uri’ parameter of the EfficientDetLiteSpec, sadly with no luck.

Anything towards the solution would be appreciated.


You can keep training the model with more data when using Model Maker. Please take a look on this thread for a workaround How to stop and resume object detector training(object detection model maker) - #14 by Viktor_Nilsson

Thank you!