DeepStream: Next-Generation Video Analytics for Smart Cities


The model used in the blog for vehicle and human classification is available for download ? or is partners model ?



We release a pre-trained ResNet-18 network for detection of
three classes of objects: car, people, and two-wheeler.

The model can be downloaded here:


Thanks Tested and working.


I wish to continue training with the weight downloaded from the link above ( But it has deploy.prototxt only. Can I download train.prototxt also?


sorry ? i don’t understand :)


I had downloaded the deep stream SDK and there are only CalibrationTable, labels.txt, deploy.prototxt and caffemodel file but no train.prototxt. Where can I get train.prototxt?


Sorry for that we don’t release the train.prototxt file for ResNet.
This example targets for demonstrating inference flow rather than training.

There are lots of information about ResNet online, and it should be easy to figure out the train.prototxt.
Thanks and sorry for the inconvenience.


could you tell us what the training data for that ResNet-18 was?

Well i just wanted to know which trainings data was used (like COCO,…), because it reaches good performance on our dataset.

But still thanks!


Could you let us know about mAP of ResNet-18 network? There is no information about its accuracy in release note or DeepStream_User_Guide.pdf.


Please check below

Hi RaviKiran,

I tried Deepstream sdk .
I was able to run it for cars.
Can I know how to run it for people. Which config file I have to use?

Also, Can I get some guidelines to extend Deepstream APIs?


The link you shared has link to caffemodel file(no deploy prototxt), where do i find the corresponding deploy prototxt for Resnet 18 pretrained on imagenet database.

Thanks !!!