PROJECT 4
Tulika Kumar
Experiment 1
Data
Face 'class_smiling_cropped\adeakin.tga' recognized as being closest too
0: class_nonsmiling_cropped\adeakin; MSE: 12971.5
1: class_nonsmiling_cropped\ddewey; MSE: 17475.8
2: class_nonsmiling_cropped\tshail; MSE: 25986.7
3: class_nonsmiling_cropped\margaux; MSE: 26480.8
4: class_nonsmiling_cropped\paullarp; MSE: 26717.1
5: class_nonsmiling_cropped\alissah; MSE: 31966.5
6: class_nonsmiling_cropped\tanderl; MSE: 39842.2
7: class_nonsmiling_cropped\crosetti; MSE: 43906.6
8: class_nonsmiling_cropped\djj; MSE: 45935.2
9: class_nonsmiling_cropped\melissa; MSE: 47657
10: class_nonsmiling_cropped\squinn; MSE: 62694.5
11: class_nonsmiling_cropped\mhl; MSE: 66077.5
12: class_nonsmiling_cropped\galen; MSE: 73924
13: class_nonsmiling_cropped\mbixby; MSE: 85813.9
14: class_nonsmiling_cropped\amiratuw; MSE:
91169.2
Face 'class_smiling_cropped\alissah.tga' recognized as being closest too:
0: class_nonsmiling_cropped\alissah; MSE: 1545.28
1: class_nonsmiling_cropped\ddewey; MSE: 12063.9
2: class_nonsmiling_cropped\margaux; MSE: 15289.4
3: class_nonsmiling_cropped\paullarp; MSE: 29846.9
4: class_nonsmiling_cropped\adeakin; MSE: 38206.2
5: class_nonsmiling_cropped\amiratuw; MSE: 43612.3
6: class_nonsmiling_cropped\djj; MSE: 54764.4
7: class_nonsmiling_cropped\crosetti; MSE: 65210.2
8: class_nonsmiling_cropped\melissa; MSE: 65691.9
9: class_nonsmiling_cropped\tanderl; MSE: 77631.6
10: class_nonsmiling_cropped\tshail; MSE: 82323.7
11: class_nonsmiling_cropped\tamoore; MSE: 108398
12: class_nonsmiling_cropped\mhl; MSE: 115041
13: class_nonsmiling_cropped\galen; MSE: 127841
14: class_nonsmiling_cropped\esp; MSE: 128921
Face 'class_smiling_cropped\amiratuw.tga' recognized as being closest too:
0: class_nonsmiling_cropped\amiratuw; MSE: 5451.23
1: class_nonsmiling_cropped\djj; MSE: 20610
2: class_nonsmiling_cropped\alissah; MSE: 29453.1
3: class_nonsmiling_cropped\ddewey; MSE: 36829.6
4: class_nonsmiling_cropped\margaux; MSE: 38969.9
5: class_nonsmiling_cropped\melissa; MSE: 42387.8
6: class_nonsmiling_cropped\adeakin; MSE: 54551.5
7: class_nonsmiling_cropped\paullarp; MSE: 63749.7
8: class_nonsmiling_cropped\esp; MSE: 67247.2
9: class_nonsmiling_cropped\mbixby; MSE: 73154.7
10: class_nonsmiling_cropped\mhl; MSE: 79840.6
11: class_nonsmiling_cropped\tanderl; MSE: 81446.1
12: class_nonsmiling_cropped\galen; MSE: 85300.7
13: class_nonsmiling_cropped\tshail; MSE: 91688.4
14: class_nonsmiling_cropped\tamoore; MSE: 111765
MISS
Face 'class_smiling_cropped\crosetti.tga' recognized as being closest too:
0: class_nonsmiling_cropped\paullarp; MSE: 11474.9
1: class_nonsmiling_cropped\crosetti; MSE: 13384.7
2: class_nonsmiling_cropped\tanderl; MSE: 38688.8
3: class_nonsmiling_cropped\margaux; MSE: 46565.5
4: class_nonsmiling_cropped\ddewey; MSE: 60211.1
5: class_nonsmiling_cropped\alissah; MSE: 60557.4
6: class_nonsmiling_cropped\squinn; MSE: 64998.9
7: class_nonsmiling_cropped\galen; MSE: 74267.5
8: class_nonsmiling_cropped\djj; MSE: 81366.9
9: class_nonsmiling_cropped\esp; MSE: 96283.5
10: class_nonsmiling_cropped\adeakin; MSE: 106503
11: class_nonsmiling_cropped\melissa; MSE: 118889
12: class_nonsmiling_cropped\amiratuw; MSE: 122194
13: class_nonsmiling_cropped\tshail; MSE: 144325
14: class_nonsmiling_cropped\mbixby; MSE: 144515
Face 'class_smiling_cropped\ddewey.tga' recognized as being closest too:
0: class_nonsmiling_cropped\ddewey; MSE: 6719.51
1: class_nonsmiling_cropped\margaux; MSE: 9926.03
2: class_nonsmiling_cropped\paullarp; MSE: 10313.4
3: class_nonsmiling_cropped\alissah; MSE: 11112.6
4: class_nonsmiling_cropped\adeakin; MSE: 21519.8
5: class_nonsmiling_cropped\djj; MSE: 26294.8
6: class_nonsmiling_cropped\crosetti; MSE: 32652.3
7: class_nonsmiling_cropped\tanderl; MSE: 34305.4
8: class_nonsmiling_cropped\melissa; MSE: 40315.3
9: class_nonsmiling_cropped\tshail; MSE: 48945.1
10: class_nonsmiling_cropped\amiratuw; MSE: 50120.3
11: class_nonsmiling_cropped\galen; MSE: 64363.8
12: class_nonsmiling_cropped\esp; MSE: 69581.4
13: class_nonsmiling_cropped\squinn; MSE: 69710.3
14: class_nonsmiling_cropped\mhl; MSE: 78876.7
Face 'class_smiling_cropped\djj.tga' recognized as being closest too:
0: class_nonsmiling_cropped\djj; MSE: 1505.53
1: class_nonsmiling_cropped\amiratuw; MSE: 20545.8
2: class_nonsmiling_cropped\ddewey; MSE: 27497.5
3: class_nonsmiling_cropped\alissah; MSE: 31505.1
4: class_nonsmiling_cropped\margaux; MSE: 33801.8
5: class_nonsmiling_cropped\esp; MSE: 38910.2
6: class_nonsmiling_cropped\paullarp; MSE: 41929.2
7: class_nonsmiling_cropped\melissa; MSE: 44160.3
8: class_nonsmiling_cropped\adeakin; MSE: 46043.2
9: class_nonsmiling_cropped\tanderl; MSE: 57457.4
10: class_nonsmiling_cropped\tshail; MSE: 68212.8
11: class_nonsmiling_cropped\galen; MSE: 69235.8
12: class_nonsmiling_cropped\mbixby; MSE: 76430.1
13: class_nonsmiling_cropped\crosetti; MSE: 79024.8
14: class_nonsmiling_cropped\mhl; MSE: 82654.1
Face 'class_smiling_cropped\esp.tga' recognized as being closest too:
0: class_nonsmiling_cropped\esp; MSE: 14089.7
1: class_nonsmiling_cropped\jwkim; MSE: 31218.2
2: class_nonsmiling_cropped\squinn; MSE: 47588.3
3: class_nonsmiling_cropped\galen; MSE: 55153.4
4: class_nonsmiling_cropped\djj; MSE: 62744.7
5: class_nonsmiling_cropped\tanderl; MSE: 64869.8
6: class_nonsmiling_cropped\mbixby; MSE: 101766
7: class_nonsmiling_cropped\rhennig; MSE: 102708
8: class_nonsmiling_cropped\melissa; MSE: 106763
9: class_nonsmiling_cropped\paullarp; MSE: 108120
10: class_nonsmiling_cropped\amiratuw; MSE: 111242
11: class_nonsmiling_cropped\margaux; MSE: 122465
12: class_nonsmiling_cropped\crosetti; MSE: 144854
13: class_nonsmiling_cropped\mhl; MSE: 148706
14: class_nonsmiling_cropped\alissah; MSE: 152659
Face 'class_smiling_cropped\galen.tga' recognized as being closest too:
0: class_nonsmiling_cropped\galen; MSE: 7017.99
1: class_nonsmiling_cropped\mbixby; MSE: 21100.1
2: class_nonsmiling_cropped\squinn; MSE: 34977.8
3: class_nonsmiling_cropped\tanderl; MSE: 52830.5
4: class_nonsmiling_cropped\esp; MSE: 70131
5: class_nonsmiling_cropped\rhennig; MSE: 75794.9
6: class_nonsmiling_cropped\jwkim; MSE: 79978.4
7: class_nonsmiling_cropped\melissa; MSE: 82495
8: class_nonsmiling_cropped\djj; MSE: 87105.1
9: class_nonsmiling_cropped\merlin; MSE: 91923
10: class_nonsmiling_cropped\mhl; MSE: 93998.9
11: class_nonsmiling_cropped\paullarp; MSE: 95291
12: class_nonsmiling_cropped\margaux; MSE: 104535
13: class_nonsmiling_cropped\adeakin; MSE: 106612
14: class_nonsmiling_cropped\ddewey; MSE: 114387
Face 'class_smiling_cropped\jaydang.tga' recognized as being closest too:
0: class_nonsmiling_cropped\jaydang; MSE: 26072.5
1: class_nonsmiling_cropped\adeakin; MSE: 31832.5
2: class_nonsmiling_cropped\melissa; MSE: 57447.3
3: class_nonsmiling_cropped\tshail; MSE: 57750.4
4: class_nonsmiling_cropped\ddewey; MSE: 62546.1
5: class_nonsmiling_cropped\mhl; MSE: 66746.2
6: class_nonsmiling_cropped\mbixby; MSE: 96871.2
7: class_nonsmiling_cropped\alissah; MSE: 99376.9
8: class_nonsmiling_cropped\margaux; MSE: 99948
9: class_nonsmiling_cropped\crosetti; MSE: 111441
10: class_nonsmiling_cropped\djj; MSE: 111569
11: class_nonsmiling_cropped\paullarp; MSE: 117774
12: class_nonsmiling_cropped\galen; MSE: 124941
13: class_nonsmiling_cropped\amiratuw; MSE: 132598
14: class_nonsmiling_cropped\esp; MSE: 146181
Face 'class_smiling_cropped\jwkim.tga' recognized as being closest too:
0: class_nonsmiling_cropped\jwkim; MSE: 5318.32
1: class_nonsmiling_cropped\esp; MSE: 36372
2: class_nonsmiling_cropped\squinn; MSE: 51489.8
3: class_nonsmiling_cropped\mbixby; MSE: 57382.3
4: class_nonsmiling_cropped\galen; MSE: 59903.8
5: class_nonsmiling_cropped\rhennig; MSE: 67378
6: class_nonsmiling_cropped\djj; MSE: 78457
7: class_nonsmiling_cropped\tanderl; MSE: 79888.8
8: class_nonsmiling_cropped\mhl; MSE: 80216.3
9: class_nonsmiling_cropped\melissa; MSE: 84405.8
10: class_nonsmiling_cropped\amiratuw; MSE: 126496
11: class_nonsmiling_cropped\merlin; MSE: 138451
12: class_nonsmiling_cropped\margaux; MSE: 144899
13: class_nonsmiling_cropped\tamoore; MSE: 147098
14: class_nonsmiling_cropped\tshail; MSE: 149874
Face 'class_smiling_cropped\margaux.tga' recognized as being closest too:
0: class_nonsmiling_cropped\margaux; MSE: 3719.92
1: class_nonsmiling_cropped\paullarp; MSE: 12796.7
2: class_nonsmiling_cropped\alissah; MSE: 14019
3: class_nonsmiling_cropped\ddewey; MSE: 29398.8
4: class_nonsmiling_cropped\tanderl; MSE: 30012.5
5: class_nonsmiling_cropped\djj; MSE: 42968.6
6: class_nonsmiling_cropped\amiratuw; MSE: 45975.4
7: class_nonsmiling_cropped\crosetti; MSE: 47608.7
8: class_nonsmiling_cropped\melissa; MSE: 47660.2
9: class_nonsmiling_cropped\adeakin; MSE: 58976.9
10: class_nonsmiling_cropped\galen; MSE: 71399
11: class_nonsmiling_cropped\squinn; MSE: 78519.9
12: class_nonsmiling_cropped\esp; MSE: 80193.4
13: class_nonsmiling_cropped\tshail; MSE: 91887
14: class_nonsmiling_cropped\mhl; MSE: 101041
Face 'class_smiling_cropped\mbixby.tga' recognized as being closest too:
0: class_nonsmiling_cropped\mbixby; MSE: 2258.52
1: class_nonsmiling_cropped\galen; MSE: 37787.5
2: class_nonsmiling_cropped\rhennig; MSE: 45390
3: class_nonsmiling_cropped\esp; MSE: 74745
4: class_nonsmiling_cropped\melissa; MSE: 75825.9
5: class_nonsmiling_cropped\squinn; MSE: 76587.4
6: class_nonsmiling_cropped\mhl; MSE: 79003.6
7: class_nonsmiling_cropped\djj; MSE: 79182.7
8: class_nonsmiling_cropped\adeakin; MSE: 81109.6
9: class_nonsmiling_cropped\tanderl; MSE: 87006.2
10: class_nonsmiling_cropped\ddewey; MSE: 96382.8
11: class_nonsmiling_cropped\jwkim; MSE: 96566.7
12: class_nonsmiling_cropped\amiratuw; MSE: 104545
13: class_nonsmiling_cropped\paullarp; MSE: 111185
14: class_nonsmiling_cropped\margaux; MSE: 115328
Face 'class_smiling_cropped\melissa.tga' recognized as being closest too:
0: class_nonsmiling_cropped\melissa; MSE: 15549.8
1: class_nonsmiling_cropped\margaux; MSE: 19682.6
2: class_nonsmiling_cropped\adeakin; MSE: 20250.4
3: class_nonsmiling_cropped\ddewey; MSE: 28311
4: class_nonsmiling_cropped\tshail; MSE: 29713.4
5: class_nonsmiling_cropped\mhl; MSE: 33875.5
6: class_nonsmiling_cropped\tanderl; MSE: 34550.6
7: class_nonsmiling_cropped\djj; MSE: 35032.2
8: class_nonsmiling_cropped\paullarp; MSE: 35709.1
9: class_nonsmiling_cropped\alissah; MSE: 36289.2
10: class_nonsmiling_cropped\squinn; MSE: 48848.8
11: class_nonsmiling_cropped\galen; MSE: 53573.1
12: class_nonsmiling_cropped\crosetti; MSE: 60178.3
13: class_nonsmiling_cropped\esp; MSE: 62472.8
14: class_nonsmiling_cropped\amiratuw; MSE: 66584.4
Face 'class_smiling_cropped\merlin.tga' recognized as being closest too:
0: class_nonsmiling_cropped\merlin; MSE: 54097.6
1: class_nonsmiling_cropped\seitz; MSE: 85327.6
2: class_nonsmiling_cropped\squinn; MSE: 220206
3: class_nonsmiling_cropped\jwkim; MSE: 233301
4: class_nonsmiling_cropped\rhennig; MSE: 242120
5: class_nonsmiling_cropped\galen; MSE: 243982
6: class_nonsmiling_cropped\mbixby; MSE: 296636
7: class_nonsmiling_cropped\tanderl; MSE: 314179
8: class_nonsmiling_cropped\esp; MSE: 316868
9: class_nonsmiling_cropped\mhl; MSE: 369628
10: class_nonsmiling_cropped\djj; MSE: 370419
11: class_nonsmiling_cropped\tshail; MSE: 373193
12: class_nonsmiling_cropped\melissa; MSE: 425440
13: class_nonsmiling_cropped\adeakin; MSE: 452167
14: class_nonsmiling_cropped\paullarp; MSE: 463344
MISS
Face 'class_smiling_cropped\mhl.tga' recognized as being closest too:
0: class_nonsmiling_cropped\melissa; MSE: 18586
1: class_nonsmiling_cropped\margaux; MSE: 19253.1
2: class_nonsmiling_cropped\alissah; MSE: 23861
3: class_nonsmiling_cropped\amiratuw; MSE: 31874.5
4: class_nonsmiling_cropped\ddewey; MSE: 39507.8
5: class_nonsmiling_cropped\adeakin; MSE: 43503.8
6: class_nonsmiling_cropped\mhl; MSE: 45869.8
7: class_nonsmiling_cropped\djj; MSE: 47391.2
8: class_nonsmiling_cropped\paullarp; MSE: 53074.3
9: class_nonsmiling_cropped\tanderl; MSE: 56036.7
10: class_nonsmiling_cropped\tshail; MSE: 75530.9
11: class_nonsmiling_cropped\mbixby; MSE: 82231.2
12: class_nonsmiling_cropped\galen; MSE: 82966.3
13: class_nonsmiling_cropped\esp; MSE: 83415.9
14: class_nonsmiling_cropped\tamoore; MSE: 93043.6
MISS
Face 'class_smiling_cropped\paullarp.tga' recognized as being closest too:
0: class_nonsmiling_cropped\alissah; MSE: 8043.42
1: class_nonsmiling_cropped\ddewey; MSE: 18170.5
2: class_nonsmiling_cropped\margaux; MSE: 30119.1
3: class_nonsmiling_cropped\paullarp; MSE: 32758
4: class_nonsmiling_cropped\adeakin; MSE: 55002.2
5: class_nonsmiling_cropped\crosetti; MSE: 58496.4
6: class_nonsmiling_cropped\amiratuw; MSE: 63450.6
7: class_nonsmiling_cropped\djj; MSE: 79203.4
8: class_nonsmiling_cropped\melissa; MSE: 96317
9: class_nonsmiling_cropped\tanderl; MSE: 99127.8
10: class_nonsmiling_cropped\tshail; MSE: 113632
11: class_nonsmiling_cropped\tamoore; MSE: 126909
12: class_nonsmiling_cropped\galen; MSE: 154036
13: class_nonsmiling_cropped\esp; MSE: 155167
14: class_nonsmiling_cropped\mbixby; MSE: 158154
Face 'class_smiling_cropped\rhennig.tga' recognized as being closest too:
0: class_nonsmiling_cropped\rhennig; MSE: 15227.6
1: class_nonsmiling_cropped\mbixby; MSE: 37176.5
2: class_nonsmiling_cropped\galen; MSE: 75458.8
3: class_nonsmiling_cropped\merlin; MSE: 86999.2
4: class_nonsmiling_cropped\jwkim; MSE: 96211.5
5: class_nonsmiling_cropped\esp; MSE: 97532.4
6: class_nonsmiling_cropped\squinn; MSE: 99232.3
7: class_nonsmiling_cropped\mhl; MSE: 124113
8: class_nonsmiling_cropped\melissa; MSE: 140080
9: class_nonsmiling_cropped\djj; MSE: 141374
10: class_nonsmiling_cropped\tanderl; MSE: 154893
11: class_nonsmiling_cropped\adeakin; MSE: 161880
12: class_nonsmiling_cropped\tshail; MSE: 176267
13: class_nonsmiling_cropped\ddewey; MSE: 196962
14: class_nonsmiling_cropped\amiratuw; MSE: 203321
Face 'class_smiling_cropped\seitz.tga' recognized as being closest too:
0: class_nonsmiling_cropped\seitz; MSE: 40058.3
1: class_nonsmiling_cropped\merlin; MSE: 50515.8
2: class_nonsmiling_cropped\squinn; MSE: 76255.6
3: class_nonsmiling_cropped\galen; MSE: 100420
4: class_nonsmiling_cropped\tanderl; MSE: 122101
5: class_nonsmiling_cropped\rhennig; MSE: 151157
6: class_nonsmiling_cropped\esp; MSE: 154275
7: class_nonsmiling_cropped\jwkim; MSE: 160881
8: class_nonsmiling_cropped\djj; MSE: 167037
9: class_nonsmiling_cropped\mbixby; MSE: 170901
10: class_nonsmiling_cropped\tshail; MSE: 173804
11: class_nonsmiling_cropped\paullarp; MSE: 183272
12: class_nonsmiling_cropped\crosetti; MSE: 187661
13: class_nonsmiling_cropped\adeakin; MSE: 212541
14: class_nonsmiling_cropped\margaux; MSE: 222984
Face 'class_smiling_cropped\squinn.tga' recognized as being closest too:
0: class_nonsmiling_cropped\squinn; MSE: 52543.6
1: class_nonsmiling_cropped\seitz; MSE: 74801.5
2: class_nonsmiling_cropped\merlin; MSE: 87844.8
3: class_nonsmiling_cropped\jwkim; MSE: 110926
4: class_nonsmiling_cropped\galen; MSE: 111005
5: class_nonsmiling_cropped\tanderl; MSE: 120372
6: class_nonsmiling_cropped\rhennig; MSE: 145225
7: class_nonsmiling_cropped\esp; MSE: 166183
8: class_nonsmiling_cropped\mbixby; MSE: 166589
9: class_nonsmiling_cropped\mhl; MSE: 216921
10: class_nonsmiling_cropped\paullarp; MSE: 220648
11: class_nonsmiling_cropped\djj; MSE: 222511
12: class_nonsmiling_cropped\tshail; MSE: 226352
13: class_nonsmiling_cropped\melissa; MSE: 234237
14: class_nonsmiling_cropped\crosetti; MSE: 240891
Face 'class_smiling_cropped\tamoore.tga' recognized as being closest too:
0: class_nonsmiling_cropped\tamoore; MSE: 9094.35
1: class_nonsmiling_cropped\tanderl; MSE: 109749
2: class_nonsmiling_cropped\mhl; MSE: 110799
3: class_nonsmiling_cropped\adeakin; MSE: 114652
4: class_nonsmiling_cropped\djj; MSE: 117339
5: class_nonsmiling_cropped\alissah; MSE: 120425
6: class_nonsmiling_cropped\tshail; MSE: 121884
7: class_nonsmiling_cropped\jwkim; MSE: 123171
8: class_nonsmiling_cropped\melissa; MSE: 130021
9: class_nonsmiling_cropped\margaux; MSE: 130358
10: class_nonsmiling_cropped\ddewey; MSE: 134084
11: class_nonsmiling_cropped\squinn; MSE: 146091
12: class_nonsmiling_cropped\paullarp; MSE: 149230
13: class_nonsmiling_cropped\amiratuw; MSE: 152695
14: class_nonsmiling_cropped\mbixby; MSE: 153428
Face 'class_smiling_cropped\tanderl.tga' recognized as being closest too:
0: class_nonsmiling_cropped\tanderl; MSE: 8774.5
1: class_nonsmiling_cropped\paullarp; MSE: 21155
2: class_nonsmiling_cropped\squinn; MSE: 31939.3
3: class_nonsmiling_cropped\margaux; MSE: 36038.9
4: class_nonsmiling_cropped\djj; MSE: 40251.3
5: class_nonsmiling_cropped\galen; MSE: 40468.4
6: class_nonsmiling_cropped\ddewey; MSE: 43697.9
7: class_nonsmiling_cropped\alissah; MSE: 46435.6
8: class_nonsmiling_cropped\crosetti; MSE: 55920
9: class_nonsmiling_cropped\adeakin; MSE: 56835.2
10: class_nonsmiling_cropped\esp; MSE: 67441.6
11: class_nonsmiling_cropped\mbixby; MSE: 67831.2
12: class_nonsmiling_cropped\melissa; MSE: 73662
13: class_nonsmiling_cropped\tshail; MSE: 74045.6
14: class_nonsmiling_cropped\amiratuw; MSE: 89548.4
Face 'class_smiling_cropped\tshail.tga' recognized as being closest too:
0: class_nonsmiling_cropped\tshail; MSE: 2435.51
1: class_nonsmiling_cropped\adeakin; MSE: 10348.2
2: class_nonsmiling_cropped\ddewey; MSE: 36649.9
3: class_nonsmiling_cropped\mhl; MSE: 40087.8
4: class_nonsmiling_cropped\melissa; MSE: 42589.7
5: class_nonsmiling_cropped\margaux; MSE: 45213.8
6: class_nonsmiling_cropped\alissah; MSE: 53722.2
7: class_nonsmiling_cropped\djj; MSE: 60015.6
8: class_nonsmiling_cropped\paullarp; MSE: 76386.2
9: class_nonsmiling_cropped\tanderl; MSE: 84556.2
10: class_nonsmiling_cropped\amiratuw; MSE: 100796
11: class_nonsmiling_cropped\crosetti; MSE: 104301
12: class_nonsmiling_cropped\jaydang; MSE: 106143
13: class_nonsmiling_cropped\tamoore; MSE: 109623
14: class_nonsmiling_cropped\squinn; MSE: 114716
Answers To Questions
1)Out of 22 people, 19 identified correctly, 3 incorrectly.
2)In the cases where identification was incorrect, the
person was listed on average as (4+7+2)/3 approximately 4th.
3)When a female face is recognized, it is not the case that
the next few best matches are females. And
when a male face is recognized, it is not the case either that the next few best
matches are male. In all the cases
where a female was recognized, there is no case where the next 4 best matches
are female. Only 1 female had the
next 3 best matches be females. For
the rest of the females that were recognized, the second and third best matches
are about evenly spread out between males and females.
In all cases where a male was recognized, there is no case where the next
4 best matches are all males. Only
2 males had the next 3 best matches be male.
For the rest of the males that were recognized, the second and third best
matches are about evenly spaced out between males and females.
4) There are tradeoffs with using a large number of users
to compute your face space. The
larger the number of users you use to compute your face space, the better your
representation of human faces. As
we have witnessed from carrying out this project, the method of using eigenfaces
to compute the face space takes minutes for less than 20 users.
If the application for which you are using a face recognition system
requires great accuracy, then you should use a larger number of training samples
to compute the face space. If your application doesn’t require great accuracy, then
you don’t need to use such a large number of training samples.
If you have a big powerful computer that can do computations really fast,
then you can use more training samples. So
in short, the answer to this question depends on the application and the time
you are willing to spend waiting for your face space to be computed.
5)The only case I can see in which it would be good to train your face space on samples other than pictures used for actual recognition is if you’re trying to determine how well your space face can generalize to new faces. Say you’ve come up with a new algorithm for computing faces spaces. If you use a set of pictures to compute this face space, your face space may be suited for recognizing only those faces it has been trained on. This isn’t a very useful algorithm then. By using samples not used in training, you can see how well the face space generalizes to new input.
Pictures Of Results
The average face - I used 25x25 pixel images (Sorry they're so small to see!).
The seven eigenfaces
Experiment 2
Data
Training Face Space on 5 faces:
Correct recognitions: 2/5 Incorrect 3/5.
Correct Recognition Rate: 40%
Training Face Space on 10 faces:
Correct recognitions: 3/10 Incorrect 7/10.
Correct Recognition Rate: 30%
Training Face Space on 12 faces:
Correct recognitions: 5/12 Incorrect: 7/12
Correct Recognition Rate: 42%
Training Face Space on 15 faces:
Correct recognitions: 6/15 Incorrect: 9/15
Correct Recognition Rate: 40%
Training Face Space on 20 faces:
Correct recognitions: 8/20 Incorrect: 12/20
Correct Recognition Rate: 40%
Answers To Questions
1) (Data for experiment shown above)
2) The incorrect recognitions were usually quite off. For example, when I computed the faces space
with 20 users and tried to recognize "adeakin", I got that the best result was "melissa" and that the
5th best result was "adeakin."
adeakin melissa
And when I computed the face space with 15 users and tried to recognize "merlin", I got that the best
result was "alissah" and that the 9th best result was "merlin."
merlin alissah
3) Here are some reasons why the program performed more poorly in recognizing people:
- Pictures of people taken in our class are (in general) very different from our undergraduate pictures.
In the undergraduate pictures people have very different facial expressions, very different angles in
which theyare facing the camera, etc. In the first experiment, the camera angles were kept the same
(the person is staring straight ahead) and the facial expressions didn't change a whole lot (a lot of smiles
are like the nonsmiles :). This is partly why we got better results in the first experiment. The matches were
more clear in the first experiment. Here they were not because of the great variability between people's
pictures in the undergraduate and class sets.
-We used less eigenfaces in our face space (or we used fewer faces from which to compute our face space).
To some extent the fewer faces you use to construct your face space, the more difficult it will be to correctly
interpolate other faces.
-We used a larger database to compute the eigenfaces from. We probably constructed face space from
people NOT is our class. So when we computed the coefficients for the people in our class, they weren't
as accurate as in experiment 1. There was probably some error in interpolating our class pictures in face
space composed from people not in our class.
4) Actually, changing the number of eigenfaces didn't change the percentage of correct recognitions much.
All of the different number eigenfaces gave about 40% correct recognition except for when I used 10 (in
which case I got 30% correct recognition). The factors I mentioned in question 3 are mostly why I didn't
see an improvement in recognition accuracy as I increased the number of eigenfaces in my face space. I got
the best recognition when I used 12 eigenfaces (42% accuracy).
5) One reason I can think of using as few eigenfaces as possible (and still get good results) is that
it is pretty time costly to compute the eigenfaces (at least with the method that we used in this project).
Pictures Of Results
The average face for 5 training samples
Eigenfaces for face space constructed from 5 training samples
The average face for 10 training samples
Eigenfaces for face space constructed from 10 training samples
The average face for 12 training samples
Eigenfaces for face space constructed from 12 training samples
The average face for 15 training samples
Eigenfaces for face space constructed from 15 training samples
The average face for 20 training samples
Eigenfaces for face space constructed from 20 training samples
Experiment 3
Answers To Questions
1) For min_scale I used 0.2. For max_scale I used 0.3. For step I used 0.05.
2) None of my cropped results get the students EXACTLY in the center. Everyone is
cut off on one side or the other a bit. But in each of the 10 cropped images I get almost all
off the person's face. See the picture results below to see what I mean.
3) If you use a min_scale that is too small, the size of your rescaled image can be
smaller than the box you use to go through and look for faces. Also, if you make your
original image really small, you lose some of its detail. A face may look nothing like a
face if the image is scaled too small.
Pictures Of Results
So here is the result of me cropping myself. As you can see the result is ok.
I got most of my face, but the right side is cut of somewhat.
Me - Uncropped (AHHH!)
Me - Cropped (AHHHHHHHHHHHHH!)
This was one of the best results I got. Just about all of Linn's face is showing...
Linn Uncropped
Linn Cropped
The cropped picture of albert below didn't turn out that well - he's missing much of
his right face!
Albert Uncropped
Albert Cropped
The one of Corey below is also one of my best results.
Corey Uncropped
Corey Cropped
I guess my cropped results weren't that good!
Experiment 4
Answers To Questions
1) For my birthday picture I used min_scale = 0.9, max_scale = 0.95, step = 0.05.
For denzel's picture, I used min_scale = 0.4, max_scale = 0.5, step = 0.05.
2) As you can see the major problem I had in detecting faces in both images is that my program
detected low-textured regions as "faces." Low-textured regions lie in the hyperplane of faces -
there are non-face things in this hyperplane. They're in the hyperplane (or close to it), but they
are far from the facial mean (as compared to actual faces). So one thing I should have done
perhaps is check the MSE of each subimage (with the average face) to see if I have something that's
even close to what an average face looks like. If it's not even close to what an average face looks
like, toss it out.
Another way perhaps to get rid of low-textured region detection would be to look at the variance of the pixels
in the subimage. If the variance is really smalll, it's probably a sweatshirt or something, not a face (it'll
be a part of the image in which all the pixels are the same color. Though most face pixels are sorta close
to one another, there's definitely more variation in the pixels of a face than say the night sky so well
found in my birthday picture. So this way I shouldn't detect the squares I ended up detecting in both
images).
Pictures Of Results
My last birthday - original
Here's the result of trying to find faces in my birthday picture:
DENZEL! - original
Here's the result of trying to find faces in denzel's picture: