Can I train additional data on my trained model?

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Hi.

I am using Deepstream 4.0.1 version and tlt-v1 on Jetson Nano.

I am training using Detectnet-v2_resnet18 in tlt-v1.

The tlt-train showed that the training time increased as the number of training data increased.

It takes a lot of time when I need to learn more data from existing data.

I don’t know much about transfer-learning-toolkit…

Is there a way to shorten the training time by learning only additional data from the model that was previously trained?

6

Yes, you can train additional data on your model which was previously trained.
For detectnet_v2 network, TLT provides pretrained model(hdf5 file) in NGC, you can set it as the pretrained model and run training.
Also, if you already trained a tlt model, you can set it as the pretrained model and run training too.

8

Hi, Morganh.

Thank you so much for the kind answer.

Could you tell me the location of the manual of the information I want?

9

See tlt user guide. Transfer Learning Toolkit — Transfer Learning Toolkit 3.0 documentation