Skip to content

Latest commit

 

History

History
14 lines (10 loc) · 934 Bytes

cascade_s2anet.md

File metadata and controls

14 lines (10 loc) · 934 Bytes

Cascade S2A-Net

Cascade S2A-Net (CS2A-Net) can be regarded as a generalized S2A-Net. It consists of several detection heads, and each head contains an Alignment Convolution and two subnetworks, .i.e, cls and reg. The overall archetecture of CS2A-Net is shown below.

The difference between CS2A-Net and S2A-Net is:

  • CS2A-Net head aligns features first by an Alignment Convolution Layer (ACL), and produce cls_score and bbox_pred later. While the FAM in S2A-Net does clssification and regression first, then produces aligned features.

  • CS2A-Net aligns features in each stage, while feature alignment in S2A-Net is only appeared at the end of FAM. So even the one stage version of CS2A-Net, .i.e, CS2A-Net-1s can produce aligned features.