Mashiro

Hi! Please follow the guide in [contributing.md](https://github.com/open-mmlab/mmengine/blob/main/CONTRIBUTING.md) to fix the lint error

Thanks for your contribution, the modification of MLFLow makes sense to me, however, we need to reconsider the modifications for the runner. I believe the core issue lies in when...

Coupling the implementation of `master_only` with the `Visualizer` is not a good practice. I believe we can avoid using the `master_only` decorator inside the `Visualizer` and instead opt for adding...

Hi, please check the modification will not break the CI

Hi! Is this PR ready for review? It remains a draft.

Hi, we should provide the OpenCV backend for `Visualizer` to draw bboxes, points, .etc, just like the show method does: https://github.com/open-mmlab/mmengine/blob/4bc2fe1aaec2d7d1464b11d50f6165494f62ef18/mmengine/visualization/visualizer.py#L228 Although there will be a lot of `if-else` logic,...

Hi, you can update the unittest in `test_visualizer.py` like this: ```python from parameterized import parameterized ... @parameterized.expand([['cv2'], ['matplotlib']]) def test_draw_bboxes(self, backend): visualizer = Visualizer(image=self.image) # only support 4 or nx4...

Thanks for your feedback I guess this is mainly caused by we've not synchronized the model buffer before saving the checkpoint and evaluating the model. We will fix it ASAP...

Maybe you could try to set: ```python model_wrapper_cfg = dict( type='...', broadcast_buffers=True, ) ``` in the config file. If it works, we'll check the implementation of `SyncBufferHook`

After configuring these parameters, did you retrain and save the new checkpoint？