Tong Gao

Could you try upgrading your MSVC to 19.xx? Our local compilation succeeded at this version.

Sorry for the confusion, I'm not an expert on Windows. 19.xx is the version of MSVC compiler instead of MSVC. Our [unit test](https://github.com/open-mmlab/mmocr/runs/7666625136?check_suite_focus=true) shows that the compilation succeeded at Windows...

It's because `mmcv-full` hasn't been installed correctly. Try the following commands: ```bash pip uninstall mmcv-full pip uninstall mmcv pip install -U openmim mim install mmcv-full ``` which should address your...

If you installed `mmocr` as a pip package, then you should prepare the configs and checkpoints following https://mmocr.readthedocs.io/en/latest/install.html#case-b-installed-as-a-package

@jiarenyf Some of the dependencies are still under construction, but they will be ready very soon.

@jiarenyf We haven't tested it yet, but it's technically feasible as long as the backbone's output matches the head's signature. Try to modify your config: ```python custom_imports=dict(imports='mmseg.models', allow_failed_imports=False) model =...

The only configuration you need is to specify `to_rgb=True` in `Normalize` transform, as done in most of MMOCR's models. Take DBNet as an example: https://github.com/open-mmlab/mmocr/blob/150cb13443134e4fd55e4f349954e024e116cbd7/configs/_base_/det_pipelines/dbnet_pipeline.py#L45-L48 https://github.com/open-mmlab/mmocr/blob/150cb13443134e4fd55e4f349954e024e116cbd7/configs/_base_/det_pipelines/dbnet_pipeline.py#L58

`NormalizeOCR` is confusing and essentially a duplicate implementation of `Normalize`. We had planned to replace them with `Normalize`, but the resulting changes were too many to be acceptable compared to...

We're also a newbie to multilingual recognition. I think this issue could be left as a general discussion.

Sometimes visualizing your model's output could be helpful