MFCFSiam: A Correlation-Filter-Guided Siamese Network with Multifeature for Visual Tracking

Figure 3

The comparison between the context padding and the zero padding in our proposed tracking framework. The tracking result is obviously not ideal. However, after the context padding, the whole target is still included in the padded bounding box, and it can still generate a significant response in the response map (maximum A). After the zero padding, the target outside the result bounding box is padded as zero. So, the maximum on the response map is compressed (maximum B). On the response map, a redder pixel represents a larger value, so maximum A is bigger than maximum B.