Some questions about MMDet 3.x visualizer

[sanbuphy]

I am trying make sense the pipeline of constructing a visualizer object. Here are my questions and thoughts.

1. What is the reason why the 'dataset_meta' needs to be used in the visualizer instance construction? Is this necessary?
   visualizer = VISUALIZERS.build(model.cfg.visualizer)
visualizer.dataset_meta = model.dataset_meta
2. add_datasample is the method of visualizer(made in mmdet/visualization/local_visualizer.py) .A visualization of an detected image itself does not need initialize dataset_meta,It can show the detected image and draw box info. My guess is that if you need to draw gt_box, then you need to get labels by dataset_meta, so we need initialize the datase_meta in the begin.
3. when i run the visualizer it will shows:"mmengine/visualization/visualizer.py:170: UserWarning: Visualizer backend is not initialized because save_dir is None. warnings.warn('Visualizer backend is not initialized '
   I think it means the visualizer need a save_dir in metaclass (MMEngine), but in mmdet/visualization/local_visualizer.py——DetLocalVisualizer class we had use save_dir: Optional[str] = None, to rewrite this method, is it necessary hint or omissions？

[ZwwWayne]

For question 1: Yes, this is necessary. The visualizer needs to obtain the class names and their color plate to show the category of boxes with consistent colors. All these information is about the dataset.
For question 2: for predictions, we also need to set the dataset meta to show the category names rather than labels.
For question 3: these might be a legacy issue to be fixed in mmdet, please @hhaAndroid to take a look

[sanbuphy]

Thank you for your reply！
For question 2, I found that it can be displayed without setting dataset meta.
For question 3, I will ask him for help, Thanks!