AP is nan #111
Description
When I use the dataset in COCO format and pretrain the model with ImageNet, all the AP values are NaN. However, when I use a COCO-pretrained model, training proceeds normally. Can you help me analyze the reason for this?
{{ message }}