deepcam-cn / face-anti-spoofing.pytorch Goto Github PK

1. I train the net with the data your provided,
2. load the checkpoint and preproess the img, output = net(x)
3. how can i get the real or fake prob with output

Thanks for your repo,
Could you share with me python version of the model?

Hi, i am curious about these two aspect:
First: I am curious about the input of net, just crop detect box of face(resize to 128*128?), is there other process?
Second: I am also curious about why net depth output hold 2 channel? depth mask label just 1 channel

Could you give me the BaiDu Drive link again? Thanks!

can you share inference code with me? many thanks

Could you reupload dataset anywhere else, baidu is too slow. Thanks

Hi, Thanks for sharing your code, could you reupload training data somewhere else? baidu is extremely slow, less than 1kb/sec. Would be nice if you can upload on google drive. Thanks

Can you give me the native .so file for arm64-v8a, please?

请问方便分享产生深度图的脚本么？

You have done a great work.
But, I have a question about the architecture about the network.
In your paper, the network details are as follows:

And I think it is similar to the network proposed in Learning Deep Models for Face Anti-Spoofing: Binary or Auxiliary Supervision.

But, when I look into your code about the definition of the network, I think there is a big difference about the implementation of the network between the code and the paper.

Or are these two networks of the code and the paper equivalent?

Looking forward to your reply.
Thanks.

I notice that the transforms for image data and depth data are not the same, there would be RandomResizedCrop for image data and just Resize for depth data. Wouldn't it lead to mismatch pixels for the generated depth and the ground truth depth?

Hello up , can you publish the so source code? Now I don't know how to implement data processing.

what's going on about the multi-frame algorithm ?

Currently I'm trying to train the network using id pro dataset, but it seems like inference is not really imrpoving, i basically got this kind of result:

threshold 0.5: frp = 0.07120253164556962 ; tpr = 0.08847184986595175

Could you suggest on how can I improve it ? Before the training I crop the faces, scaled it to 256x256, labeled it correctly, used your dataset as an example. Your is training pretty well

Hi, Thanks a lot for this. Could you please however let know why the 'part of the training' data Baidu link is dysfunctional? The link does not return any file.

Regards

首先非常非常感谢,作者的开源精神,再次感谢.
我运行了安卓的demo,效果是很好的,也看了一下输出的信息,想不明白为何要keep static 呢? 感觉安卓demo和训练的模型运行机制不一样啊,
我训练了模型,也理解了作者的意图,引入了depth 信息,但是测试的时候效果不及安卓demo,还请大佬指点一下

Thanks for great work. Can you show me how I can convert this to Flask project and have it running as an API thank you.

hi,thanks for sharing~ if I trained my own model ,how to convert it into android project and how to do inference in android? thanks~

What is graphic card you used for training ? I do have RTX2060 SUPER and seems like its not enough :(
I do receive the following:

RuntimeError: CUDA out of memory. Tried to allocate 512.00 MiB (GPU 0; 8.00 GiB total capacity; 5.92 GiB already allocated; 107.97 MiB free; 171.50 KiB cached)

Hi!

Thanks a lot for this repository!

Is it possible to download the data without baidu account? I can't create one from Russia((((

Thanks for your repo, I saw that training set using depth modal, so does test phase need depth model?

Could you also upload depth generators for training set. I'd like to test on another dataset also. Later I can share model here also

I don't find any license information. Can I use this source code in my application? Is there any limitation?