How to enable dynamic input shapes in TF-TRT

Hi, I have a tensorflow model with dynamic input shapes, i convert the model by tf-trt with ‘is_dynamic_op=True’. However, as my input images have different shapes, one TRT engine will be built for each input image.
Then i found this pr:, it seems that a single engine can handle input with different sizes through dynamic input shapes (supported since TensorRT version 6).
How can i use it in tf-trt? Thank you very much.