An Evaluation of Deep Learning Methods for Small Object Detection

Figure 4

The location of the default boxes in different scales: (a) image with GT boxes; (b) 8 × 8 feature map; (c) 4 × 4 feature map.

(a)

(b)

(c)