Zaida Zhou

Hi @levan92 , yet, the `petrel_client` is a backend for internal use. In fact, we have a plan to support a file client for S3 object storage next month. Of...

> Sure, will an implementation that utilises `boto3` work? sure

The issue will be resolved by #1545

Hi @hoya-cho , you can calculate the loss in the method (https://github.com/open-mmlab/mmdetection/blob/3e2693151add9b5d6db99b944da020cba837266b/mmdet/core/evaluation/eval_hooks.py#L117) and add it to key_score.

> I know that the loss value is not returned in **single_gpu_test** or **multi_gpu_test** when in val mode. How can I calculate the losses by taking them..? > > `result...

Hi @innerlee , you can merge the upstream master to your branch to fix the error of building documentation

https://mmdetection.readthedocs.io/en/latest/1_exist_data_model.html  Can we use the `merge_from_dict` to implement the same behavior like `--cfg-options checkpoint_config.max_keep_ckpts=1` https://github.com/open-mmlab/mmcv/blob/44e7eee835c3bc138ee0f667228777eca3db1a17/mmcv/utils/config.py#L550

> * For simple replacement, yes. But the --cfg-options thing is verbose and hard to use > * For complex cfgs such as multiple references, the --cfg-options is not powerful...

We also need to update the usage of Config (https://mmcv.readthedocs.io/en/latest/understand_mmcv/config.html) and add some unit tests for the feature.

Hi @innerlee , is there any progress?