choyingw / SynergyNet

3DV 2021: Synergy between 3DMM and 3D Landmarks for Accurate 3D Facial Geometry
MIT License
382 stars 57 forks source link

evaluation of AFLW dataset #21

Closed yjwnet9 closed 2 months ago

yjwnet9 commented 2 years ago
캡처 캡처

Face alignment evaluation result of AFLW in the paper and my evaluation result are different.

Could you provide the evaluation code of AFLW dataset?

choyingw commented 1 year ago

self.keypoints = _load(osp.join(d, 'keypoints_sim.npy'))

This line extracts commonly used 68 point landmarks for AFLW2000-3D eval. AFLW usea 21 point landmarks, and from their landmark definition converting directly from 68 points would cause large error. We picked the indices that best match AFLW's definition (https://drive.google.com/file/d/1AYRUrrV_Kc8sa50EHmd-Sc8QmiVflCU0/view?usp=share_link). Just replace the mentioned line with this npy file and do the evaluation.

facial_landmarks_68markup

Screenshot from 2022-12-16 23-36-44