Re-train network with new labeled frames

[v738301]

Hi! @davidhildebrand @spoonsso @data-hound

I am now trying to re-train my previously refined network with more labeled data from other animals (The experiment setups are the same). What is the proper way to do this?

Should I use finetune mode to retrain my model with more data? (I seems that every time you finetune a network, it add a new layer to it) or there are other way to do this?

Thank you!

[data-hound]

Hi @v738301

As of now, finetuning the model with new data is the only way to get the model learn from newer frames. I would not recommend doing it multiple times.
About adding an extra layer, there is a fix for that which is being tested currently and will be released soon (sometime after the new release) - which is supposed to resolve this issue.

For now, I would recommend to do a finetune once you have all the labeled data