Minghui Liao

I suggest you comment it on the PR which enables the Docker. [here](https://github.com/MhLiao/TextBoxes_plusplus/pull/76)

I will check the model.

Do you use GPU mode or CPU mode? There is a bug in the CPU code. I have fixed the bug in CPU code now. The results are normal in...

You can get the char boxes from `char_mask` like [this](https://github.com/MhLiao/MaskTextSpotterV3/blob/9bd595e40c693ced0dc8ce4f33d65048424189e5/maskrcnn_benchmark/engine/text_inference.py#L391). The `char_mask` can obtain like [this](https://github.com/MhLiao/MaskTextSpotterV3/blob/9bd595e40c693ced0dc8ce4f33d65048424189e5/maskrcnn_benchmark/engine/text_inference.py#L86).

@HHeracles Right

@ouyangshizhuang 有过实验，但是效果不佳就没有采用。

@yyr6661 Try to set `CHAR_MASK_ON` as `False` to disable the character segmentation branch. Also, you need to change the number of character classes, rewrite the [char2num](https://github.com/MhLiao/MaskTextSpotterV3/blob/b7114ef9b7c76050c81deab6674d99e02e4aef1c/maskrcnn_benchmark/utils/chars.py#L7) and [num2char](https://github.com/MhLiao/MaskTextSpotterV3/blob/b7114ef9b7c76050c81deab6674d99e02e4aef1c/maskrcnn_benchmark/utils/chars.py#L17). I think...

> @MhLiao If I train on Chinese and turn of char mask, should I change the predictor with no "Char" in roi_mask_predictors.py? e.g. SeqMaskRCNNC4Predictor. Because predictor with "Char" will use...

@HHeracles Did you use the suggested thresholds for different types of the lexicon in the evaluation scripts? The results you show are slightly lower, which may be caused by the...

@HHeracles Have you followed the pretraining and finetuning schedules as suggested in the paper? In my opinion, you may need more iterations and a smaller initial learning rate if your...