sai-bi / FaceAlignment

Face Alignment by Explicit Shape Regression
MIT License
340 stars 205 forks source link

How much time does it cost to do predicting for an image? #16

Open wguo68 opened 9 years ago

wguo68 commented 9 years ago

Can the algorithm predict landmarks in 15 ms as the paper? Thanks

sai-bi commented 9 years ago

It takes about 10ms.

sai-bi commented 9 years ago

You may also refer to this issue https://github.com/soundsilence/FaceAlignment/issues/17

wguo68 commented 9 years ago

It cost 160ms in our tests,what't the problem?

On Sunday, May 10, 2015, Bi Sai notifications@github.com wrote:

You may also refer to this issue #17 https://github.com/soundsilence/FaceAlignment/issues/17

— Reply to this email directly or view it on GitHub https://github.com/soundsilence/FaceAlignment/issues/16#issuecomment-100577337 .

sai-bi commented 9 years ago

In fact, in the test demo program provided by me, the initial number is set to 20, that is, each image is predicted 20 times, and then output the average as final output. Of course this will be much slower. In practice, you may change the initial number to a small number such as 2 or 3, it will be much faster and the accuracy will not be affected too much.