Does converting all operations to onnx->tensorrt speed up a lot?

[YoungjaeDev]

Hello

https://github.com/obss/sahi , This GitHub is for Tilling Inference.

I'm developing it in two directions.

1. deploy c++ with tensorrt (I have completed some development, and the result is the same, but the speed is faster than pure python)
2. (slice-letterbox-inference-nms-boxmerge) As you pursue in this repo --> What I'm curious about is, let's say all operation onnx conversion is possible, but when I convert it to tensorrt and perform it with one engine, will the speed exceed number 1 (deploy c++) ? I haven't tried it either, but I'm curious based on your experience.

I think if it were faster, there would be someone who would try like this.