Table 1 Comparisons of averaged errors of our MSRRN with different face alignment methods on the 300-W dataset, where 68 landmarks were employed for evaluation

FPLL (Zhu and Ramanan 2012)

8.29

8.16

8.22

18.33

10.20

DRMF (Asthana et al. 2013)

6.57

6.70

6.65

19.79

9.22

RCPR (Burgos-Artizzu et al. 2013)

6.56

5.93

6.18

17.26

8.35

GN-DPM (Tzimiropoulos and Pantic 2014)

5.92

5.69

5.78

–

–

SDM (Xiong and la Torre 2013)

5.67

5.50

5.57

15.40

7.50

CFAN (Zhang et al. 2014)

5.44

5.53

5.50

–

–

ERT (Kazemi and Sullivan 2014)

–

–

–

–

6.40

BPCPR (Sun et al. 2015)

–

–

5.24

16.56

7.46

ESR (Cao et al. 2012)

–

–

5.28

17.00

7.58

LBF (Ren et al. 2014)

–

–

4.95

11.98

6.32

LBF fast (Ren et al. 2014)

–

–

5.38

15.50

7.37

Deep reg (Shi et al. 2014)

–

–

4.51

13.80

6.31

CFSS (Zhu et al. 2015)

4.87

4.63

4.73

9.98

5.76

CFSS prac (Zhu et al. 2015)

4.90

4.72

4.73

10.92

5.99

TCDCN (Zhang et al. 2016)

4.57

4.60

4.80

8.60

5.54

MSRRN

3.98

3.71

3.83

7.25

4.84