Can the Zed 2 utilize TensorRT gpu acceration for speeding up Yolov4 inference? My main concern with buying a zed is that the gpu acceration won’t be fast enough without tensorRT as I need to run by object detection in realtime with a decent fps.
Yes, the object detection module is using TensorRT for inference, and the upcoming release with custom model input will allow you to use any inference code you want, including TensorRT such as this tensorrtx/yolov4 at master · wang-xinyu/tensorrtx · GitHub
Thank you, when will the upcoming release come out?
It should come out this week
Thank you. And just to confirm. Custom models are supported by the Zed 2 correct? Not only the Zed 2i?
Yes, ZED2, ZED2i and from today also ZED Mini.
The new SDK v3.6 has been released today
Hi~You mean I can use the zed2 camera + my own trained maskrcnn model for identifying wood end faces, not just the model used to identify people in the official sample program?
Exactly, you can find documentation, tutorial, and examples here:
Thank you very much！
excuse me~I I have another question. I see the example on the official website, zed2 + maskrcnn, which can measure the distance and detect objection at the same time, but the example of detect objection on the official website only shows the detection box. Can zed2 + yolo5 detect the objection and measure the distance at the same time?
Sure. Look at this part of the code: