search

ami-iit / onnx-cpp-benchmark

Simple tool to profile onnx inference with C++ APIs.
BSD 3-Clause "New" or "Revised" License
1 stars 1 forks source link

Missing network warm-up #11

Open giotherobot opened 1 year ago

giotherobot commented 1 year ago

The benchmark does not perform any warm-up of the gpu in case it is used, so the results will be incorrect.

traversaro commented 1 year ago

Probably it is sufficient to skip counting the time for the first iteration (see https://github.com/ami-iit/onnx-cpp-benchmark/blob/main/onnx-cpp-benchmark.cpp#L366).

giotherobot commented 1 year ago

Yes indeed!