Can I use dlib package to extract 68-point landmarks?
Posted by: SYSU_LuoJ @ March 17, 2021, 8:42 a.m.Hi, We already provide the 5-landmarks by MTCNN. Is this information OK for you? Please make sure.
I alread have the five point extrated by MTCNN，but I need a more precise landmarks。
Posted by: SYSU_LuoJ @ March 17, 2021, 1:35 p.m.Hi, you may have a try by including your own face landmark detector in your submission. I don't know if dlib can work in this way, but you can also try other detectors like here: https://github.com/ainrichman/Peppa-Facial-Landmark-PyTorch
For the method to include extra packages in your submission, see our last post. It is better to first test your code locally using our docker image.
If all you trials failed and you still must use 68 landmarks, please report back, and we may consider adding the 68-landmarks to our API in the next detection phase (begins Mar. 29).
Hi, we have added the 68-landmark information in our json file. You can use it now. Please refer to updates of the starterkit github.
Posted by: bob_peng @ April 12, 2021, 4:19 a.m.