Distance Field-Based Convolutional Neural Network for Edge Detection

Table 4

The statistic comparison with some competitors on BSDS500 dataset.

Method	ODS	OIS	FPS
Canny [7]	0.611	0.676	28
EGB [23]	0.614	0.658	10
MShift [24]	0.598	0.645	1/5
gPb-UCM [25]	0.729	0.755	1/240
Sketch Tokens [26]	0.727	0.746	1
MCG [4]	0.744	0.777	1/18
SE [9]	0.743	0.763	2.5
OEF [27]	0.746	0.770	2/3
DeepContour [16]	0.757	0.776	1/30†
DeepEdge [17]	0.753	0.772	1/1000†
HFL [28]	0.767	0.788	5/6†
N4-Fields [29]	0.753	0.769	1/6†
HED [18]	0.788	0.808	30†
RDS [30]	0.792	0.810	30†
CEDN [31]	0.788	0.804	10†
MIL + G-DSN + MS + NCuts [32]	0.813	0.831	1
RCF [19]	0.806	0.823	30†
RCF-MS [19]	0.811	0.830	8†
Experiment 4.2.1	0.802	0.818	17†
Experiment 4.2.2	0.813	0.831	17†
Experiment 4.2.3 (DF-CNN)	0.818	0.833	16†
PiDiNet [20]	0.807	0.823	92†
PiDiNet-L [20]	0.800	0.815	128†
PiDiNet-Small [20]	0.798	0.814	148†
PiDiNet-Tiny [20]	0.789	0.806	152†

† means GPU time.