1 paper · 1 filter
Gary Leung, Jun Gao, Xiaohui Zeng +1
Existing transformer-based image backbones typically propagate feature information in one direction from lower to higher-levels. This may not be ideal since the localization abilit…