Data structure in Menpo2D_semifrontal_train.txt

[js19710423]

First, thanks for your interesting contribution!

I've downloaded Menpo2D_semifrontal_train.txt and I've seen that it contains a set of 75 points. The last 68 correspond to the facial landmarks, but I am not sure about what are the initial 7 points.
The four corners of the face rectangle and something else?
Thank you in advance.

[kadirbeytorun]

I think 2 of those 7 are bounding box for the face and 5 are seperate landmarks found by retina-face.
I am not sure about the order though, maybe the first two= bounding box

[js19710423]

Thank you so much!!

[ignaciohrdz]

Hi, I just had the same question today and I can confirm the order:

1. Top-left corner of the bounding box (x, y)
2. Bottom-right corner of the bounding box (x, y)
3. Left eye (x, y)
4. Right eye (x, y)
5. Tip of the nose (x, y)
6. Left corner of the lip (x, y)
7. Right corner of the lip (x, y)
8. The remaining facial landmarks (either 68 or 39)

This is what I did to get the actual facial landmarks:

with open("Menpo2D_semifrontal_train.txt") as f:
    lines = [l.rstrip() for l in f.readlines()]
    images = [l.split(" ")[0] for l in lines]  # image names
    landmarks = [l.split(" ")[1:] for l in lines]  # points (x y x y x y...)
    facial_landmarks_x, facial_landmarks_y = landmarks[14::2], landmarks[15::2]

[starhiking]