alexanderparkin/casia-surf_cefa

Face Anti-spoofing Attack Detection Challenge@CVPR2020

casia-surf_cefa's Issues

casia SURF test.zip password.

I have downloaded the casia surf dataset but I am unable to unzip the test set. I tried casia-cefa-test-572121-usd89f but it says incorrect password.

I would greatly appreciate if someone can provide the password for test.zip

thank you.

rachel

data

Hi, everyone!
Could you share the dataset with me, please?
my email: [email protected]

There is no data/OpticalFlow.cpp file in this repository

As the title, the file didn't be put on the repository.
I saw the comment on README.md. The modification of the file is about removing logs spamming to console.
I try to remove them by myself. However, I found there are some error logs during training phase.

"All the pixels are invalid in estimation Laplacian noise!!!
Something severely wrong happened!!!"

Are these error logs OK for this project?

Thank you for taking the time to read this issue.

Have you tried to use the module of "RGBVideoWrapper"?

Hi, this work is cool.
I notice that you have used a transformer in your experiments, but I did not find it in your paper.
Could you tell me the results of your trial of using the transformer?

about data

hello, thanks for your work.
now, i have the CeFA data, i know the password of test.zip, but i do no thave the unzip password of dev.zip and train.zip , can you tell me the password?
Looking forward to your raply

cuDNN error occured

Thank you. We are getting a lot of help from your resources. I followed the guidelines given in your source. However, after a while, an error occurs suddenly.

It has RTX 3080 and CUDA's version is 11.

RuntimeError: cuDNN error: CUDNN_STATUS_EXECUTION_FAILED
Any solution?

Inside at_learner_core/trainer.py, forcibly passing torch.device to "cpu" will start training without problems. But it will be slow, right? Can't I use it normally while using gpu?
Should I downgrade the CUDA version? But does the RTX 3080 maintain compatibility with versions below CUDA 11? We ask for your opinion.

Can we get Pre-Trained Model

same as the title.

想问一下 您rankpooling的操作是怎么实现的呢

The labels of test file

Dear. Sir:
I have check the devtest.txt, but I found the label of test parts are "-1". Could your please offer the true label of test data?
Thank you very much

Best

WeiHL

how to predict a new picture with pretrained model?

真实与攻击的图片处理方式为什么不同?

在查看数据集发现,真实数据进行了人脸检测并裁剪,而攻击却未进行人脸检测裁剪, 数据预处理方式不同,难道不影响预测结果跟实际应用么? 那些方方正正人脸大的图片容易预测为真,而那些东倒西歪,一部分像素才有人脸的图片容易预测为假?

请问能不能分享一下数据集呢

比赛已经结束了，希望可以分享一下数据集，万分感谢作者！

Are you possible from python code to c++ code for prediction ?

If you can convert it to c++, I can pay for it. Of course, it would be a reasonable cost.
It may sound a little rude, but I mean it.
Please contact me; [email protected]

Optical flow cannot run GPU

I join github of pyflow and see that it hasn't had GPU yet. It's make this program quite slow.
So why you choose this, and does you have any replace solution for Optical flow to make this faster.

Metrics calculation

Each video was an object in the metrics calculation, i.e. one video clip == 1 tp/fn/etc, am I right?

ERROR: Unexpected segmentation fault encountered in worker.

Dear Alex. I followed your instructions to perform the necessary preparations. But when I execute this piece of code an error occurs.
CUDA_VISIBLE_DEVICES=0 python main.py --config experiments/rgb_track/exp1_protocol_4_1/rgb_track_exp1_protocol_4_1.config;

This is the complete error. I'm guessing there is something wrong with ransform, but I don't know how to fix it. Can you help me?