Self-Recurrent Learning and Gap Sample Feature Synthesis-Based Object Detection Method

Table 4

Mean average precision experimental results on the VOC2017 dataset.

Model	Backbone	Input size	[email protected]
Faster R-CNN [21]	VGG-16	1000 × 600	73.20
SSD513 [43]	ResNet-101	513 × 513	80.60
HyperNet [9]	VGG-16	1000 × 600	76.30
YOLO [44]	GoogleNet	448 × 448	63.40
Ours	Darknet-53	416 × 416	88.70

We compared object detection results on four state-of-the-art methods to demonstrate the effectiveness of our method.