search

chengzhengxin / sdm

Supervised Descent Method Apply to Face Alignment, and Head Pose Estimation with Linear Regression. It is cross-platfrom, easily compile in windows, ubuntu, even in Android & iOS.
448 stars 197 forks source link

training format #30

Open zhanglaplace opened 6 years ago

zhanglaplace commented 6 years ago

hi，i change the training code by write my own function the landmakr format is x1,x2,,,,x68 y1,y2,,,,,y68. after 5 stage my training error from 1.2--->0.05 with normalized but when i test the result with my own model, only in the first frame has a correct landmark ,and soon all landmarks disappeared.. what's my problem....

zhanglaplace commented 6 years ago

add test result and error

xiexiexxs commented 6 years ago

hi, i met the same problem.

when tested on each single image ,the landmarks is right. when tested on video,landmarks on the first frame were ok, while landmarks on subsequent frames are wrong and disappeared quickly.

i find that the landmark detection on the first frame is based on box by the face detector while those on on subsequent frames are based on boundingboxes of landmarks on the previous frame. the face boxes or the initial shapes are quite different, which leads different landmarks.

so have you solved the problems, and how?

many thx.