chelsea234 / multi-domain-learning-fas Goto Github PK

Hi,

thank you very much for your work and for releasing the code :) I have read your code carefully but still have some doubts. How do you prepare datasets? And how can I get frames from videos?

Looking forward to your reply....

Seems there is typo with the arrange of the model initialization.

When I construct the FASMD dataset based on the list (SIW E train and OULU E train) you provided, I found that I got the E sub-dataset with 839(from siw)+1620(from oulu) videos, which is a lot more than the 1696 listed in the paper?

Hi, in the inference.py file, I see that you load all of models in a list (model_list, model_p_list, model_d_list), but I couldn't see anywhere all of lists can be used in the inference.py file. Can you help me to clear this point to me, please?

Hi
Please add and a requirements file to define version of the required library for inferencing.
when I run the inference file I faced this error:

face-alignment is not defined

Hi,
can i have the siw-m dataset link? thanks!

Hi, i have sign DRA form and send email to [email protected] since 7/11

But i haven't get access to dataset, can you check that

Hello,
I sent the request form and the signed DRA to ask for the download link of the SiW dataset a month ago but I haven't received any reply so far. I also sent you an email a few days ago asking for help. If possible, can you check my email and help me with downloading the dataset.
Thank you very much.

Hi @CHELSEA234 ,

Thanks for directing the link. I am trying to run the script for the inference : source/test_architecture.py

In this file, it requires trained model with 2 other folders of RE & G_Op as written here : https://github.com/CHELSEA234/Multi-domain-learning-FAS/blob/main/source/test_architecture.py#L68

Please guide me on this

Thanks

can you share you datasets

Hi, I don't understand what is the purpose of recon of the liveness image, is it possible or can it be used to obtain kind of a numerical output that states whether an image is live or spoof?

What is the Release # in this DRA form on top-right?

Thanks!

Hello, first thank you for this great work. Below is the error I encounter when running source_SiW_Mv2/preprocessing.py,

It seems that the "eye2eye_dis" of the current frame equals to zero, which also makes "xr-xl" equal to zero. The x_scale then cannot be compute (since it will be infinite). May I ask is there the same problem happends when you run this code? I wonder if it is ok to ignore the frame directly. Hope to here from you soon. Thank you very much.

Hi @CHELSEA234

When I run the inference.py file using this command:

python3 inference.py --cuda=0 --pro=1 --dir=./demo/live/ --overwrite --weight_dir=./resources/save_model_siwmv2_pro_1_unknown_Ob

but it gives an error :

File "inference_ed.py", line 194, in test_step
    img, img_name = dataset_inference.nextit()
  File "project/Multi-domain-learning-FAS/source_SiW_Mv2/dataset.py", line 121, in nextit
    return next(self.feed)
  File "torch_env/lib/python3.8/site-packages/tensorflow/python/data/ops/iterator_ops.py", line 816, in __next__
    raise StopIteration
StopIteration

How can I solve this issue?

In the Paper, you use TPR@FPR=0.5% as metrics but this metric is not included in metric.py (But I find the TPR@FPR=0.2%, 5%, and 1%). And I am curious about the test_architecture.py because it can not work well for the function test_update not being used.

hi, I see the model outputs depth, region, content, and additive traces, how is it possible to transform this to live/spoof to replicate your paper?

Thanks for sharing the code. I'm new to FAS and would be very grateful if you could give me some data preprocessing details.

1. According to the paper, the FAS model and framework are frame-based, right? In preprocessing, is it necessary to extract frames from the video? Then crop out the face from each frame? How many frames are extract from each video?