If I first clone and build jetson-inference:master, and install it, and then clone jetson-utils:master and install it, the jetson-inference library breaks.
(There are missing symbols – apparently something got re-factored but not merged back?)
Hi snarky, yes the jetson-inference:master branch should be used with the jetson-utils branch that it’s submodule is currently on ( 798c416).
The jetson-inference:dev branch is up-to-date with jetson-utils:master if you want to try it. I have been doing some upgrades, and it should be backwards-compatible with it’s public-facing APIs, but let me know if something breaks. There were some minor functions renamed/refactored in jetson-utils but those weren’t commonly used externally.
Thanks, Dusty! I had started using some of the color corretion functions from jetson-utils before I realized the jetson-inference version was older, so the linker errors came from my code.
On another note: Is there any chance we’ll see a uff version of Yolo or TinyYolo versions 3 or 4 anytime soon?
If you are doing a bunch of color conversions, you might want to check out the new cudaConvertColor() function, which automatically picks the conversion routine based on the input/output image formats (see the imageFormat enum).
I’m not sure, I think the YOLO models are sticking with ONNX for now. There was a TensorRT sample added under /usr/src/tensorrt/samples/python/yolov3_onnx
That was exactly the function I tried using :-)
Thanks for the info.