I have got the transfer learning toolkit up and running. I would like to know which example in the jupiter notebook will show me how to train a model that will equal the performance of :
“Deepstream-app -c source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt”
Hi,
For TLT, there are 4 Jupyter notebooks inside the docker: classification,detectnet_v2,Faster-rcnn and SSD.
Each of them can run their corresponding pre-trained models which are available at ngc.nvidia.com.
For DS,please check the model file via your mentioned configuration file of Deepstream.
So resnet-10 seems to be the model for :
Deepstream-app -c source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt
Do I basically pick any one of these:
classification,detectnet_v2,Faster-rcnn and SSD.
To retrain resnet-10 to my images for my application?
Hi
What will your application plan to inference, for classification or for detection?
For classification,please select classification pre-trained model.
For detection,please select Faster-rcnn,SSD or detectnet_v2.
Which type of detection did you guys use:Faster-rcnn,SSD or detectnet_v2 to get Resnet-10 to perform at the speed and accuracy that the nano is able to accomplish using:
Deepstream-app -c source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt
Hi adventuredaisy,
For the resnet-10 model inside DS, please refer to below tickets to get more information.
[url]https://devtalk.nvidia.com/default/topic/1063932/deepstream-sdk/which-resnet-detector-is-this-/[/url]
I suggeset you can train all the three networks with your own dataset.