Gender recognition from frontal facial images within project FP7-ICT-247525-HUMAVIPS

Tunning size of base window for SVM OCAS solver

<< zpět

Overview

The best dimension of base window
Fig.1 The best dimension of base window showed on ten random images

The output of face detector is cropped face from image, but this cutout don't has to be the best input for gender classifier. The aim of this experiment is to find out seemingly the best base window size when classifier reach the best results.

Note: In future experiments will be used dimension 60x40 (the yellow one) regardless of results of this experiment. Old experiments calculated with this dimension so in new experiment we going to use this dimension to keep independence of experiments.



Results

Parameters of experiment
imSize [h x w]80 x 64
C0.01
win positioncentered
nPyramids4
winSize [h x w]Validation error
40 x 3010.47
42 x 3010.63
44 x 3010.61
46 x 3010.45
48 x 3010.21
50 x 3010.15
52 x 3010.72
54 x 3010.08
56 x 3010.12
58 x 309.89
60 x 3010.31
62 x 309.87
64 x 309.94
66 x 309.89
68 x 3010.30
70 x 3010.28
40 x 3210.37
42 x 329.73
44 x 329.54
46 x 329.49
48 x 329.77
50 x 329.36
52 x 329.79
54 x 329.68
56 x 329.66
58 x 329.57
60 x 329.71
62 x 329.52
64 x 329.66
66 x 329.63
68 x 329.91
70 x 329.86
40 x 3410.00
42 x 349.73
44 x 3410.10
46 x 349.59
48 x 349.91
50 x 349.49
52 x 349.66
54 x 349.66
56 x 349.70
58 x 349.66
60 x 349.64
62 x 349.49
64 x 349.49
66 x 349.86
68 x 349.59
70 x 349.79
40 x 369.52
42 x 368.98
44 x 369.26
46 x 368.71
48 x 369.59
50 x 368.91
52 x 369.47
54 x 368.82
56 x 369.01
58 x 368.76
60 x 368.91
62 x 369.06
64 x 368.96
66 x 368.98
68 x 369.13
70 x 369.20
40 x 389.42
42 x 389.45
44 x 389.42
46 x 389.31
48 x 389.12
50 x 389.31
52 x 389.40
54 x 389.33
56 x 389.20
58 x 389.35
60 x 389.45
62 x 389.26
64 x 389.26
66 x 389.61
68 x 389.73
70 x 389.91
40 x 409.33
42 x 408.89
44 x 409.49
46 x 408.75
48 x 409.26
50 x 408.54
52 x 409.06
54 x 408.78
56 x 408.89
58 x 408.80
60 x 408.89
62 x 409.20
64 x 409.08
66 x 408.96
68 x 408.94
70 x 409.45
40 x 429.33
42 x 429.12
44 x 429.06
46 x 429.01
48 x 429.40
50 x 428.68
52 x 429.31
54 x 429.12
56 x 429.27
58 x 429.08
60 x 429.22
62 x 429.12
64 x 429.38
66 x 429.26
68 x 429.47
70 x 429.64
40 x 449.27
42 x 449.06
44 x 449.29
46 x 448.57
48 x 449.08
50 x 448.68
52 x 448.98
54 x 448.66
56 x 448.94
58 x 449.06
60 x 448.99
62 x 448.89
64 x 449.24
66 x 449.13
68 x 449.43
70 x 449.31
40 x 469.05
42 x 469.29
44 x 469.15
46 x 469.19
48 x 468.98
50 x 469.20
52 x 469.50
54 x 469.19
56 x 468.89
58 x 469.29
60 x 469.47
62 x 469.15
64 x 469.27
66 x 469.63
68 x 469.70
70 x 469.59
40 x 489.42
42 x 489.31
44 x 489.24
46 x 489.17
48 x 489.27
50 x 488.89
52 x 488.99
54 x 489.06
56 x 489.12
58 x 488.91
60 x 489.24
62 x 489.59
64 x 489.40
66 x 489.47
68 x 489.49
70 x 489.89
40 x 509.10
42 x 509.06
44 x 508.96
46 x 509.20
48 x 509.24
50 x 508.87
52 x 509.08
54 x 509.27
56 x 509.33
58 x 509.06
60 x 509.22
62 x 509.40
64 x 509.31
66 x 509.40
68 x 509.36
70 x 509.89
Testing error for minimal validation error: 24.96%
<< zpět