How does the pytorch version test GFLOPs and FPS, author

[2308700388]

Star RTDETR 请先在RTDETR主页点击star以支持本项目 Star RTDETR to help more people discover this project.

---

Describe the bug A clear and concise description of what the bug is. If applicable, add screenshots to help explain your problem.

To Reproduce Steps to reproduce the behavior.

[lyuwenyu]

You can use torch.profiler

https://pytorch.org/tutorials/recipes/recipes/profiler_recipe.html

[2308700388]

Hello author, I understand what you mean, but I used the model variable in the model to test the model of the model on the 3090 device, the input is (1,13,640,640), and the gflops of the test R18RTDETR is 30, I don't know if this is probably correct or not

[2308700388]

I just don't know how long it takes to calculate FPS and GFLIPs in RTDETR plus post-processing process

[lyuwenyu]

1. RTDETR-R18 has 60 FLOPs
2. only need once forward pass

[20231211]

@lyuwenyu 请问RTDETR-R50的FLOPs是88G吗

[lyuwenyu]

见主页文档 @20231211

[syl07061]

Dear Author,

I downloaded the rtdetr_r50vd model you provided to calculate the GFLOPs with an input of (1, 3, 640, 640). The result I obtained was only 69 GFLOPs, which is significantly different from the 108 GFLOPs mentioned on your homepage. Why is there such a discrepancy? What was the input data size used in your tests? I look forward to your response.