Ayush Thakur

Hey @levan92, is the issue that the WandbLoggerHook is not logging validation metrics or is it that the steps (x axis) are not correct (meaningful). Also since you are using...

Hey @levan92, I have faced the same issue but didn't dig deeper. I have a hunch that it has something to do with MMDetection's `workflow`. The `MMDetWandbHook` subclasses MMCV's `WandbLoggerHook`...

> based on wandb's documentation, it increments wandb's step by 1 every time it is called. Therefore, when it is validation's turn to log, wandb will be ahead of the...

Thanks for checking it out @levan92. I will investigate more in this direction and make a PR to fix it.

I am trying to achieve the same thing. While investigating, I realized that the `WandbLoggerHook` methods have access to the `runner` object. However, the `runner` has no `cfg` class variable....

For now, I am doing this to log config to W&B: ``` cfg.log_config.hooks = [ dict(type='WandbLogger', init_kwargs={'entity': 'ayut', 'project': 'mmdetection-test', 'config': cfg._cfg_dict.to_dict()}, interval=10, ignore_last=True, reset_flag=False, commit=True, by_epoch=True, with_step=True, log_artifact=False)] ```

Hey @zhouzaida, would love to know what you think about it. I can make a PR if you would like.

Hey @chunxt will you be able to share a colab or a GitHub gist to reproduce the issue?

Hey @ConvMech if you would like to make a PR, I would happily review it. :)

Great pointers by @bhack. To add two cents to this exact philosophy - @soumik12345 and I are working on implementing YOLOv2 (to learn and share mostly but also come up...