the onnx inference scores is different with pth inference

WongKinYiu / yolov7

Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors

GNU General Public License v3.0

13.27k stars 4.19k forks source link

the onnx inference scores is different with pth inference #577

Open lxzatwowone1 opened 2 years ago

lxzatwowone1 commented 2 years ago

I train a single class model ，the pth inference is correct ， the onnx inference boxes almost but the scores different with pth inference， look like half but closely not half，this is my conver command python3 yolov7/export.py --grid --simplify help me！！！

opassos commented 2 years ago

I found the same issue when exporting to onnx. The scores appear to be roughly half

mnurilmi commented 2 years ago

@lxzatwowone1 is this problem clear? I mean, I'm having the same problem

lxzatwowone1 commented 2 years ago

Sorry I couldn't log in to github some time ago，the problem may be that fp16 is used，please change use fp32