ZQCNN icon indicating copy to clipboard operation
ZQCNN copied to clipboard

关于106landmark的问题

Open guoguoguilai opened this issue 5 years ago • 4 comments

如果要使用您提供的精确106点模型,需要自己完成图片裁剪吗?

guoguoguilai avatar Apr 26 '19 08:04 guoguoguilai

看我的SampleMTCNN里怎么用的

zuoqing1988 avatar Apr 26 '19 08:04 zuoqing1988

速度真的是超快,但是感觉精准度不太够,想知道在训练过程中有加入heatmap的学习吗?

guoguoguilai avatar Apr 26 '19 09:04 guoguoguilai

就是个简单的回归。如果要更准,可以自己弄点数据用train-mtcnn训练。

zuoqing1988 avatar Apr 26 '19 09:04 zuoqing1988

https://github.com/aidlearning/AidLearning-FrameWork/tree/master/examples/landmark106 Fast:already converted your models to ncnn,so it is fast and multi-threads. 15 fps running on the mobile with NO tracking! using python code in the android mobile phone

aidlearning avatar May 04 '19 04:05 aidlearning