yolov8s在960x960输入size存在较大概率误检，同一个模型用onnx或者torch推理正常的

wang-xinyu / tensorrtx

Implementation of popular deep learning networks with TensorRT network definition API

MIT License

6.99k stars 1.77k forks source link

yolov8s在960x960输入size存在较大概率误检，同一个模型用onnx或者torch推理正常的 #1426

Open asflower opened 10 months ago

asflower commented 10 months ago

Env

GPU, e.g. V100, RTX2080, TX2, Xavier NX, Nano, etc. GTX1660S、RTX2080Ti
OS, e.g. Ubuntu16.04, Win10, etc. Ubuntu1804
Cuda version cuda-11.4
TensorRT version 8.5

About this repo

which branch/tag/commit are you using? master
which model? yolov5, retinaface? yolov8s

Your problem

1、yolov8s，960x960的模型，当输入图像小于960x960时，较大概率一些类别存在误检情况; 2、同一个模型，改成640x640输入，误检消失; 3、同一个模型、同一批图像，用onnx或者torch推理正常，没有误检;

wang-xinyu commented 10 months ago

@lindsayshuo 帮忙确认下呢

lindsayshuo commented 10 months ago

训练的参数我看看

asflower commented 10 months ago

不好意思，我这边对模型训练不是太了解，这里只做工程部署。我怀疑问题可能出在编译的engine模型上，之前也尝试过替换前处理，问题还是存在

stale[bot] commented 8 months ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

RaidenCJ commented 6 months ago

你好，请问找到原因了么？方便共享下解决方案么 @asflower 我也遇到了这个问题

lindsayshuo commented 6 months ago

已解决，看最新代码

stale[bot] commented 4 months ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.