search

koushiksrivats / FLIP

Official implementation of the paper "FLIP: Cross-domain Face Anti-spoofing with Language Guidance". (ICCV 2023)
https://koushiksrivats.github.io/FLIP/
65 stars 3 forks source link

Frame generated with MTCNN #3

Open hamhanry opened 1 year ago

hamhanry commented 1 year ago

Hi, thank you for your amazing work.

Currently i would like to reproduce the results. I notice that during my dataset preprocessing, if the selected frame using the protocol stated

only sample two frames: frame[6] and frame[6+math.floor(total_frames/2)]

what if on selected frame, the MTCNN do not detect any face, what is your approach in that scenario?

As your code required to have frame0 and frame1. thanks

koushiksrivats commented 1 year ago

Hi Thank you for your interest in our work.

Do you have a rough count for how much such selected frames in each dataset, the faces are not there ?

hamhanry commented 1 year ago

Hi @koushiksrivats i notice during producing the frame 1, several video just could not detect any face by using the MTCNN, some could not detect face due to bluriness, some just because of the face are not there.

In your approach, do you implement the the extraction from each frame or you just directly use the chosen label that generated from the SSDG generate label function here?

koushiksrivats commented 1 year ago
  1. Which MTCNN implementation are you using ?
  2. Could you try splitting the videos into frames and using MTCNN on each frame, and finally choosing frame[6] and frame[6+math.floor(total_frames/2)] ?
hamhanry commented 1 year ago

Dear @koushiksrivats

  1. The MTCNN is already using this implementation https://github.com/YYuanAnyVision/mxnet_mtcnn_face_detection
  2. yes, i just trying to re-run the face detector that the face images in casia especially in _HR format video, somehow always no face found. So whats exactly your treatment? just increase the frame to find any face?

thanks

hamhanry commented 1 year ago

Dear @koushiksrivats

  1. The MTCNN is already using this implementation https://github.com/YYuanAnyVision/mxnet_mtcnn_face_detection
  2. yes, i just trying to re-run the face detector that the face images in casia especially in _HR format video, somehow always no face found. So whats exactly your treatment? just increase the frame to find any face?

thanks

and to be more precised, its only happening in Casia, the rest dataset are okay.