faceAlignment_tensorflow icon indicating copy to clipboard operation
faceAlignment_tensorflow copied to clipboard

face alignment of state-of-art implemented with tnsorflow

Results 9 faceAlignment_tensorflow issues
Sort by recently updated
recently updated
newest added

您好,请问如何自己生成meanshape数据 The mean shape refers to the average of aligned shapes for training faces, where alignment is done by fitting with similarity transformation. 这个是原文,只有这一句话,然后没有查到其他资料

![image](https://user-images.githubusercontent.com/26001162/64671364-cd8af980-d49a-11e9-881a-519f63d9b80e.png) 好像效果很差……。打扰您了,请问这是什么原因呢?

('valid num ', 554) 2019-07-08 21:45:57.680160: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA 2019-07-08 21:45:57.900107: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1411] Found device 0...

首先感谢分享您的代码,在您给的源码中,图像和关键点的矫正使用到了TransformParamsLayer这个函数,这个函数的输入是S0和Stage1的输出,也就是说人脸位姿的矫正是依赖标定的关键点的,我想问的是,这样做会不会太依赖标注的关键点了?实际应用时可没有标注好图像。在做测试时,没有了标注的关键点进行位姿矫正,得到的结果是不是很差?我在demo_dan.py测试了你给的几张图片,测试的结果不怎么好。

release里只找到了dan的预训练模型,请问eval_prnet 中256_256_resfcn256_weight 在哪里可以找到啊

stacked-HG原文中是每个HG结构中间输出结果计算loss,在您的复现中没有看到这个结构,只有一个output返回,请问是不是应该所有output联合计算loss?

best_fit_rect 函数中 fitted_shape += box_center - target_center 在求fitted_shape时target_center应该还是要*scale吧~