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

<div>Heatmap visualization examples of our proposed method MF<sup>3</sup>M and other current fusion methods embedded in RetinaNet. <i>f</i><sub><i>i</i></sub>, <i>i</i> = 1, 3, 5, means the output feature of <i>i</i>-th level in pyramid network.</div>

Computational Intelligence and Neuroscience

fig5

Figure 5

Figure 5: A Novel Pyramid Network with Feature Fusion and Disentanglement for Object Detection 