ak-nv
commented
3 years ago Currently, TLT model can be deployed with DeepStream and not supported for jetson-inference
Once you get etlt model; you can use DeepStream as suggested in Readme: Perform inference using DeepStream SDK on Jetson
Please let me know if you have questions for deploying it using DS.
Hi, thanks again for all your help, I've been trying really hard to make this work but I have a couple issues, when I run the code in the workstation using a webcam, I get fantastic FPS, however even under great lighting conditions it behaves in a funny way, It sort of waits (I don't know if it's like buffering) for a minute, then it shows a ton of inferences all at the same time, like "no mask" with a ton of red frames, then no inference at all, then again a lot of inferences all at the same time.
Also I tried deploying it to the jetson nano (because I thought this behaviour could somehow using the model in a deployment), I got all the way up to getting the trt file and the fp16 engine, but after checking a lot of documentation, and all the versions and conflicts I ended up so very confused, most of the posts I find seem outdated, and not compatible with for example, tensorflow 2.x,
I would like to deploy the trained model using the jetson.inference module in python, in order to integrate it with the jetson GPIO, but I just can't find a concrete example, is there any chance you could redirect me to what I have to read/study in order to do this? Something like this: https://github.com/dusty-nv/jetson-inference/blob/master/docs/detectnet-example-2.md ? Cheers!