Training and testing procedures are not the same

[wuqiangch]

trafficstars

I have a question.In the training, the results of the previous level of training are not used for the latter level of training. However, in the test phase, the previous results are used as a reference for the face position of next test phase. My understanding is like this. Is it right?

[luoyetx]

What if level-1 is failed, the data processed by level-1 are not stable, there might be some wrong data for level-2. Randomly sample around a point can cover the situation that a landmark is far away from the ground truth. However, if a landmark from level-1 is failed (too far away from the ground truth)m level-2 can not help.

[wuqiangch]

I have trained the nets several time and always got worse a liitle than your model's result . Just because of the network initialization?

[luoyetx]

maybe you should run more iterations or tuning the learning rate and other parameters during the training goes on.