search

protossw512 / AdaptiveWingLoss

[ICCV 2019] Adaptive Wing Loss for Robust Face Alignment via Heatmap Regression - Official Implementation
Apache License 2.0
395 stars 88 forks source link

Metrics for COFW dataset, Inter-ocular or Inter-pupil? #4

Closed BeierZhu closed 4 years ago

BeierZhu commented 4 years ago

Hi, In section 7.2, you announced to use inter-pupil distance for normalization ("For the COFW dataset, we use inter-pupil (distance of eye centers) as the normalization factor". However, in Table 2, the metric for other previous SOTA methods(Robust face landmark estimation under occlusion, etc,) used inter-ocular distance as the normalization factor. Is this a mistype?

protossw512 commented 4 years ago

@BeierZhu Hi, it is not a mistype, I actually followed an ECCV 2018 paper: A Deeply-initialized Coarse-to-fine Ensemble of Regression Trees for Face Alignment(Please refer to table 3) to use inter-pupil distance as the normalization factor.

I also noticed some metrics listed on my table are actually inter-ocular normalized, such as Disentangling 3D Pose in A Dendritic CNN for Unconstrained 2D Face Alignment, I will correct those ones and provide my inter-ocular normalized NME as well on arXiv.

Please note that using inter-ocular distance as the normalization factor will actually result in a smaller NME, so the inter-pupil normalized NMEs will always larger than the inter-ocular normalized NMEs.