Some questions about Pascal Voc Evaluation index

Thank you for your work, I would like to know that in your work, Table 7 is about the performance on VOC. May I ask whether these evaluation indicators, such as AP and AP50, are calculated according to the COCO dataset？
When I independently trained RetinaNet-RestNet50 on VOC dataset without distillation, if according to the evaluation index of COCO, AP only reached 54.1, but AP50 could reach 82.4, I am quite confused about this. Could you please give me some suggestions.

No files, tests, or entry points are identified. Start by checking the evaluation setup for VOC and COCO and the reported AP and AP50 values; done means documenting which metric definitions and dataset conventions produce the results in Table 7.

Written by the indexing model from the issue text.