Thanks to visit codestin.com
Credit goes to Github.com

Skip to content

long inference time of using TensorrtExecutionProvider #97

@noahzn

Description

@noahzn

Hi, since we can not guarantee that the same numbers of keypoints are extracted and matched, using TensorrtExecutionProvider will generate cached files for the first time and these caches will be used later for faster inference. However, when the network meets some new inputs that are not in the optimized ranges of cached files, the engine will generate new cached files and this may take more than 1 minutes. This is very dangerous as we are running the network online. Do you have any idea how we can solve this issue? I have also raised a similar issue in the ONNX repo. microsoft/onnxruntime#22822

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions