How can I use the tao-toolkit to train additional differences to an existing model?
For example, I would like to add our original dataset to “model.tlt” in the following URL and generate a new trained file.
My original dataset is the annotated images and class files.
This looks like more related to the TAO toolkit we are moving this post to the TAO forum to get better help.
You can use the peoplenet model as the pretrained model and then train with your own data.
For example, a pretrained model is a file named .tlt.
Suppose that the pretrained model can only detect people.
I want to add a newly annotated cat image and a class file to it and generate a file that can distinguish between people and cats.
human + cat .tlt
Suppose the time spent learning the cat took 3 hours.
Then I want to be able to detect people, cats and dogs again.
Suppose the dog for learning also took 3 hours.
I want to be able to spend only 3 hours in this case.
is it possible?
Can you please tell me the URL of a document that might be helpful?
Yes, a good pretrained model can play this important role. You can find some info in Preparing State-of-the-Art Models for Classification and Object Detection with NVIDIA TAO Toolkit | NVIDIA Technical Blog