* Update action.md

delete the step of change model name of fight recognition

* Update video_action_infer.py

get model path from model_dir without model name

* optimize action vis