An Agile and Efficient Neural Network Based on Knowledge Distillation for Scene Text Detection

Table 4

Comparison results on MSRA-TD500. “,” “,” and “” represent “Precision,” “Recall,” and “-score,” respectively. The best and second-best terms are highlighted in italic and bold, respectively.

Method	(%)	(%)	(%)
DeepReg [30]	77	70	74	1.1
RRD [31]	87	73	79	10
MCN [32]	88	79	83	—
PixelLink [13]	83	73.2	77.8	3
Corner [33]	87.6	76.2	81.5	5.7
TextSnake [15]	83.2	73.9	78.3	1.1
CRAFT [28]	88.2	78.2	82.9	8.6
DB-ResNet-18 (512) [1]	85.7	73.2	79	82
DB-ResNet-18 (736) [1]	90.4	76.3	82.8	62
DB-ResNet-50 (736) [1]	91.5	79.2	84.9	32
The proposed (Pr0.5) (512)	85.8	73.5	79.2	95
The proposed (Pr0.5) (736)	87.2	79.9	83.4	64