li SegLink:网络结构 conv4-conv11间的尺寸依次减少（每一层是前一层的 1/2),从6个特征图上检测多尺度的Segm㎡ent和Link 1=1 1=2 1=3 1=4 1=5 1=6 crocs crocs crocs s Clipes crocs Combining Segments 3x3 conv predictors crocs VGG16 i through convl crocs 4.3 conv7 pool5 te刀 conv 64 32 82 conv 16 92 conv 10.2 com 11 512 1024 1024 512 256 256 Input Image (512x512) 1024.k3s1 256,k1s1 128.k1s1 128.k1s1 256,k3s2 Detections 1024.k1s1 512k32 256.k32 256.k32 借鉴SSD思路，采用VGG16作为backbone进行特征提取SegLink:网络结构 借鉴SSD思路，采用VGG16作为backbone进行特征提取 conv4-conv11间的尺寸依次减少（每一层是前一层的 1/2），从6个特征图上检测多尺度的Segment和Link