yolov7.onnx give me lower mAP results

[ydgoo9]

Hello, I am exporting the yolov7.pt to yolov7.onnx for inference on OpenCV 4.6.0 with C++.

I could make a yolov7.onnx successfully like below.

python export.py --weights cfg\deploy\yolov7.pt --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640 --device 0

I build the cpp app with yolov7.onnx and could detect the object.
I found my app could not detect some object but they are detected on python with original.

Can you recommend the guideline for it ?