While the
classifier is still not perfect, it is a dramatic
improvement over the initial result!
The
Key :
Augmenting
the training data with synthetic faces.
Best
Result for Original Training Data
|
Best
Result for Augmented Training Data
|
|
|
Asian |
0% |
Black |
13% |
Hispanic |
0% |
White |
97% |
(Neural
Network)
|
Asian |
69% |
Black |
47% |
Hispanic |
0% |
White |
85% |
(Multi-Class
GLD)
|
|