For example, I have two trained models of the same neural network.
They both have a “.tlt” file extension.
I want to combine them to generate a new “.tlt” file.
How can I do that?
A.tlt → pretrained cat detection model
B.tlt → pretrained dog detection model
A.tlt + B.tlt = C.tlt
→ tao toolkit learning models
→ new .tlt file that can discriminate between cats and dogs
It is still necessary to train a new model with these two kinds of dataset.
Does that mean I can’t create a model that can detect new objects from a pre-trained model?
For example cat.tlt and dog.tlt.
Combine them and create cat+dog.tlt (a model that can detect cats and dogs)
No, you can train a model that can detect new objects from a pre-trained model. But there is not a way to combine them directly. To create cat+dog.tlt, it is needed to prepare cat and dog dataset. The cat.tlt or dog.tlt can be a good pretrained model during the training.