Transfer learning problem

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hi，

Deepstream SDK has resnet10.caffemodel,how to use this model for transfer learning.I have my training server Tesla V100 card, and I also have the tlt-streamanalytics container.

thanks.

2

Hi Kevin,
TLT is a tool for transfer learning. It enables end user to fine-tune pre-trained models with their own data.
But user cannot use tlt to transfer learn DS resnet10 caffe model directly.

The tlt pre-trained model at ngc.nvidia.com is not a copy of DS resnet10 caffe model.

TLT can help end user train their own data and get tlt model, prune the tlt model, retrain the pruned tlt model, export to etlt model or generate a trt engine for deploying in Deepsteam.

3

Thanks Morganh，

 Which container of NGC can directly train the Caffe model of resnet10, and then verify it by DS test ？
4

Hi Kevin,
TLT container is designed to help customer train their own data. It is close to your request although
it cannot transfer learn DS resnet10 caffe model directly.
You can prepare your own data and select detectnet_v2 network inside tlt container to train, then prune/retrain/export and verify in Deepstream.