Data Set Description For XM2VTS
There are total 2360 faces in the XM2VTS data set.
All the data set are manually labeled with 68 points which be called the face landmarks point.
XM2VTS data set is the ideal face data set to testing your ASM/AAM fitting algorithm so far. Original image size is 720x576. In our ASM/AAM fitting we reduce it to 360x288. During our test, we use Viola Jones algorithm to detect the face. But there are two face 338_1_2.jpg and 340_1_1.jpg that can't be detected by all 4 haarcascade_frontalface.xml files.
Our ASM/AAM Fitting Result
We use xxx_1_1.jpg as the training set (295 faces). After ASM/AAM training, we fitting all the XM2VTS data set (2630 faces). Our best result is that the average point error is ave_err = 0.342344 for ASM fitting. Here is how we define the ave_err: for each face i (i=1..2358 faces, excluding 338_1_2.jpg and 340_1_1.jpg since can't detected face); do the ASM fitting, compare each point (68 points) with the manually landmarked point; calculate the average_error_i, i.e, sqrt(sum(calculated_value - manually_landmark_value)^2)/68; and then average the error, ave_err=sum(average_error)/2358; We believe our result is best result for ASM so far. If you can achieve a better result, let us know and we'll pay you $1.
Our Face Top 5 Match Result
How to use AAM do the face match is an interesting topic.
Althrough there are some papers talk about using AAM do the face recognition, but we really don't see some good results.
Here is what we did:
Using AAM to extract each face feature which is a low dimension (168 data).
Then use the face feature data to do the mapping. For each face in XM2VTS, find the top 5 high similarity match.
The speed is pretty good; It only cost 110ms to match one face against 2358 faces.
The match rate is amazing; It is 98.81% (for each face, if we can find the same person excluding itself on the top 5, we take it as a success).
ASM/AAM Fitting Notation
- ASM1/xxx_x_x.jpg is the ASM fitting result with triangle mask.
- AAM1/xxx_x_x.jpg is the AAM fitting result with outline mask.
- AAM2/xxx_x_x.jpg is the AAM fitting result with triangle mask.
- AAM3/xxx_x_x.jpg is the AAM fitting result with appearance.
- AAM5/xxx_x_x.jpg is the AAM fitting result with appearance on the mean shape with outline.
- AAM7/xxx_x_x.jpg is the AAM fitting result with alignment to the mean shape with outline.
- AAM8/xxx_x_x.jpg is the AAM fitting result outline mask.
- AAM9/xxx_x_x.jpg is the AAM fitting result triangle mask.
