BigDong

did not get your points, more details are needed

https://github.com/open-mmlab/mmdetection/blob/master/docs/en/tutorials/config.md the main codes are in configs/_base_, for example: https://github.com/open-mmlab/mmdetection/blob/56e42e72cdf516bebb676e586f408b98f854d84c/configs/_base_/schedules/schedule_1x.py#L11 https://github.com/open-mmlab/mmdetection/blob/56e42e72cdf516bebb676e586f408b98f854d84c/configs/_base_/datasets/coco_detection.py#L32

what command and config did you use? Please use issue template

Seems this is a pycocotools error, have you checked your data annotations?

Kindly ping @chhluo have a look

what checkpoints did you use?

Thanks for your reply, we will fix ASAP

In mmdet, samples per gpu * GPUs = batch size epoch is training epoch, that you are setting total_iteration per epoch = total_sample / batch_size

seems this is an annotation error, please check whether your annotation is correct.

mmdet did not support this, but I think maybe you can use the mask image to deal with the original image. because usually mask image is tensor.bool(or np.bool) and the...