Perception Encoder: The best visual embeddings are not at the output of the network

Benchmark Model Rank Results
object-detection-on-coco-minivalPE_spatial (DETA)#2box AP: 66.0Params (M): 1900