* 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