Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Inference the STARK-Lightning onnx models with ONNXRunTime(C++) #82

Open
Stoooner opened this issue Apr 1, 2022 · 8 comments
Open

Inference the STARK-Lightning onnx models with ONNXRunTime(C++) #82

Stoooner opened this issue Apr 1, 2022 · 8 comments

Comments

@Stoooner
Copy link

Stoooner commented Apr 1, 2022

这边已经将STARK-Lightning的onnx模型以ONNXRunTime(C++)版本进行了推理, 在Quadro P4200显卡上能够跑到170 FPS左右, 在Jetson Agx Xavier这种边缘计算设备上能够跑到80~90 FPS左右. 并且在性能更加羸弱的jetson TX2(大疆 妙算2)上测试了速度能够达到30FPS多。我的邮箱是[email protected], 有需要探讨的可以先给我发邮件,我git不一定常看这些消息.

@ChenYang-1996
Copy link

请问您的项目地址?

@TsingWei
Copy link

lightning的效果跟本没法落地

@penghouwen
Copy link
Member

lightning的效果跟本没法落地

Really? There are many ways to optimize a research model for practical deployment. Looking forward to seeing your advance.

@TsingWei
Copy link

lightning的效果跟本没法落地

Really? There are many ways to optimize a research model for practical deployment. Looking forward to seeing your advance.
因为Lightning模型设计上并没有输出目标的置信度,哪怕是像ST那样有一个score模块都能勉强用。对于实际使用来说,Lightning这样的设计在找不到目标时只会胡乱给一个框,从模型训练的角度来看,它必然会给出一个尽可能大的框来获得更小的损失,而这种设计对落地来说毫无意义。当然在有限的场景下,Lightning优化速度的设计还是比较不错的。

@Stoooner
Copy link
Author

Stoooner commented May 6, 2022

请问您的项目地址?

我上传了我自己的GitHub,只不过暂时处于private状态(因为项目里用到了这部分代码),如果你确有需求可以私聊我或者留下你的微信我加你,谢谢。

@Stoooner
Copy link
Author

Stoooner commented May 6, 2022

必然

确实,lightning的效果确实如你所说,我这边也的导出了stark_st50的onnx并且以onnxruntime(c++)进行了推理,这个效果不错。但是就如你所说的,lightning用于性能比较羸弱的计算设备上是比较合适的,而stark_st50导出的onnx在性能羸弱的设备上速度就比较差了。此外我还在性能更差的Jetson TX2上测试了lightning(onnxruntime c++)的代码,能够达到30fps以上,因此效果还行。我在着手将stark的onnx以TensoRT(c++)的方式进行推理并且辅以cuda加速前后处理,等有了结果我再提issue.

@Stoooner
Copy link
Author

Stoooner commented May 6, 2022

lightning的效果跟本没法落地

Really? There are many ways to optimize a research model for practical deployment. Looking forward to seeing your advance.

确实,lightning的效果确实如你所说,我这边也的导出了stark_st50的onnx并且以onnxruntime(c++)进行了推理,这个效果不错。但是就如你所说的,lightning用于性能比较羸弱的计算设备上是比较合适的,而stark_st50导出的onnx在性能羸弱的设备上速度就比较差了。此外我还在性能更差的Jetson TX2上测试了lightning(onnxruntime c++)的代码,能够达到30fps以上,因此效果还行。我在着手将stark的onnx以TensoRT(c++)的方式进行推理并且辅以cuda加速前后处理,等有了结果我再提issue.

@Stoooner
Copy link
Author

我这边把tensorRT版本的跑起来了,在另外一个issue中,各位有兴趣可以去看看

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants