TrafficCamNet or VehicleTypeNet for Vehicle Classification


I am working on Jetson Nano 4GB module and the application is Vehicle Classification. As I can see on NGC some pretrained models are available:

So I want to train the above model with custom dataset. Which is more preferable for our use case with good accuracy, good FPS and low resource utilization?

Looking forward for your valuable comment on this.

Vikas Dwivedi


It should depend on your use case.

VehicleTypeNet is a ResNet18 based classifier.
TrafficCamNet backbone is a DetectNet_v2 detector with ResNet18 as a feature extractor.

So please check if you need a detector or not first.

For example:
To apply a classifier on an ROI region, please use TrafficCamNet.
To apply the classifier to the whole image or other detector’s output, you can use VehicleTypeNet.