A Novel Pyramid Network with Feature Fusion and Disentanglement for Object Detection

Figure 7

Different positions of deformable convolution in MF³F. The capital D stands for deformable convolution.

(a)

(b)

(c)