Other DETRs with PositionRelationEmbedding

We integrated our PositionRelationEmbedding into other DETR variants, here we provide some of their pretrained weights. The weights are named using the following pattern:
{model}_{backbone}_{img_size}_{dataset}_{training_setting}.pth
The enhanced models is named with ++ to distinguish them from their original versions.

Relation-DETR

Relation-DETR checkpoints for object detection, weights are named using the following pattern:
{model}_{backbone}_{img_size}_{dataset}_{training_setting}.pth