search

ZhangYuanhan-AI / CelebA-Spoof

[ECCV2020] A Large-Scale Face Anti-Spoofing Dataset
531 stars 92 forks source link

方便提供深度图么？ #27

Closed 22wei22 closed 3 years ago

22wei22 commented 3 years ago

请问方便把数据集对应的深度图提供一下么？ 请问训练代码准备开源么？

ZhangYuanhan-AI commented 3 years ago
  1. 我们会陆续整理，但是深度图的生成方法好像不难。
  2. 训练代码我们可能没办法完全开源，因为公司的关系。但是我们会尽量回答所有训练细节。
22wei22 commented 3 years ago

请问您方便分享一下得到人脸深度图的代码么？我使用的这个项目，https://github.com/clks-wzz/PRNet-Depth-Generation不过我的一些图片检测不到人脸 请问这个中心点还进行偏移，是什么含义 image

我的邮箱是1334637558@qq.com

eesaeedkarimi commented 3 years ago

Based on the answer of the owner here the input is the cropped face. Therefore I think you can to replace retinaface with dlib in the PRNet and crop the images then use PRNet to find the depth map. retinaface itself can choose the center image. If you want to use the CelebA-Spoof they provided bbox of faces and there is no need to use retinaface to find bboxes. Just follow the instruction in the dataset to crop the images.