mediapipe or face alignment when extract lm2d

[fff-ttt]

您好，我在extract_lm2d.py中，看见您写了弃用face alignment的原因。 我在提取2d face landmark时候，对比使用mediapipe和face alignment，发现：

1. mediapipe提取速度快，但总是遇到“未检测到landmarks”的问题
2. face alignment很少遇到“未检测到landmarks”的问题，但提取速度奇慢 不知道你们当时有没有遇到类似的问题，如果遇到了是怎样解决/平衡的呢？

非常感谢！

[Hina-Anwar]

I am facing failed to detect ldm issue, any idea how to resolve it?

Warning: failed detect ldm in idx=5818, use previous frame results.
Warning: failed detect ldm in idx=5831, use previous frame results.
Warning: failed detect ldm in idx=5836, use previous frame results.
Warning: failed detect ldm in idx=5844, use previous frame results.
Warning: failed detect ldm in idx=5850, use previous frame results.
Warning: failed detect ldm in idx=5854, use previous frame results.
Warning: failed detect ldm in idx=5881, use previous frame results.
Warning: failed detect ldm in idx=5896, use previous frame results.
Warning: failed detect ldm in idx=5941, use previous frame results.
Warning: failed detect ldm in idx=5943, use previous frame results.
Warning: failed detect ldm in idx=5944, use previous frame results.
Warning: failed detect ldm in idx=5945, use previous frame results.