I tried the way you said, it worked. Im facing issue in live detection if i use web cam for live capture and detection my detection window is not opening proper, i’ll share you the picture of that. kept trying different approachs but im not getting what i want.
that is most probably issue with the code. I can’t comment much with these details without looking at code. You are not getting proper output window while using webcam, but are you getting it in other cases like using offline video?
I don’t think the would work on offline video, seems like post-processing is wrong, you are not doing dfl in post-processing, which was removed from model to make it convertible to rknn (if you had done it). Your inference code for image might follow a completely different path. You can’t completely rely on AI generated code to understand all the modifications you have done and implement accordingly.
what you can try is make a function which takes image as input and returns annotated (detection drawn) image. And, when that works just tell the AI to use this function to extend it to video display.
cant we use yolov26 ? because it works better than other models.
yes, it is possible to run yolo26 on axon. we had tried that.
ok, The process after onnx conversion is same like yolov8 ? eg,head removal,post process etc ?
yolo26_2rknn.py (11.8 KB)
yolo26_inf.py (9.9 KB)
can check these scripts out and modify according to your requirement