Yolov8 convert issue

[WillWarti]

Hi, I would like to draw to your attention to the following convertors behaviour : I trained a Yolov8 nano version model from Ultralytics SDK, I have no problem when I convert it to ncnn format directly from this framework and deploy it in a C++ application : input is a 640x640 image and output is a 8400x7 matrix (3 classes only) as expected. But when I convert it firstly to ONNX from Ultralytics and then convert to NCNN format using NCNN convertor (20240201 version) or online convertor, I noticed that output size is wrong in C++ application (8400x38 matrix). See attached both .param models. Sincerely yours.

yolov8_corrupted.param.txt yolov8_from_ultralytics.param.txt