When converting your XML format to COCO, I have found some issues. This is not meant to be criticism, I just thought you might want to know if you are not aware of it. Thank you very much for publishing this amazing dataset!
the "freight_car" class is referred to as "feright_car", "feright car", "feright".
I have also found a "truvk"
One annotation has the name "*".
There are local Windows paths in the annotation files that don't have anything to do with the naming scheme of the images. Only the name of the xml files counts.
There are xml files for images that do not exist.
Some images (examples: trainimg/09012.jpg, trainimgr/16009.jpg) are truncated
When converting your XML format to COCO, I have found some issues. This is not meant to be criticism, I just thought you might want to know if you are not aware of it. Thank you very much for publishing this amazing dataset!
trainimg/09012.jpg,trainimgr/16009.jpg) are truncated