How we improve Tensorflow API models performance on Docker

I am trying to put in production my tensorflow models for image porcessing (like Yolo for exemple), so to that purpuse we did transform our models to API, then we created an image docker where the APIs models will be running.

The problem is that the performance of our API models that are running on Docker image are very slow, around 1s per frame/image, and the performance of our models without the Docker is much faster around 0.01s per frame/image.

Is there a way to improve the performance of our API models while running on the Docker.

Our docker is running on a graphic card TITAN RTX with 24G

Source: Docker Questions