Rotary Position Embedding for Vision Transformer

Benchmark Model Rank Results
object-detection-on-coco-val2017Swin-B + RoPE-Mixed (DINO, 12ep)#120box AP: 54.5
object-detection-on-coco-val2017ViT-L + RoPE-Mixed (DINO-ViTDet, 12ep)#159box AP: 52.9
object-detection-on-coco-val2017ViT-B + RoPE-Mixed (DINO-ViTDet, 12ep)#192box AP: 51.2