Dose not need to pull the tao docker image from ngc?

I view how to use the tao toolkit to train custom model at https://developer.nvidia.com/tao-toolkit.
I have not find the step to pull the tao image container, just see a docker login step.
I want to ask a question is : only pip install the nvidia-tao package and dose not need to pull the tao docker image from ngc?

thank you！