About inference time and deployment

[Vonct]

Hi, I'm now trying to deploy the model. But it turns out that the inference time is about 230ms. (On TRX2060). Comparing to the result that you mentioned on the paper, it's not normal. So I want to know what should I be careful about the deployment.
What's more, I find that the preprocess takes also a long time. (About 120ms). How can I improve it?
Thanks in advance.