Train data format

[vedrusss]

Thank you for your nice implementation and share!

Could you please give an example of your data (face boxes and landmarks) provided for CelebA?
I wounder the data format:

1. does face box represented as or ? In absolute or ratio coordinates (ratio to image width, height)?
2. do facial landmarks coordinates are presented as X1 Y1 X2 Y2 ... X5 Y5? Or X1 X2 X3 ... Y3 Y4 Y5? In absolute or ratio coordinates?

Next, CelebA stores 'regression_nums_' (random numbers) into 'rectangle.txt' (line 157: txt_write(regression_nums_, "rectangle");). While 'generate_hdf5.py' needs 'regression_box.txt'. Is that just filename confusion? Do I really must provide random numbers instead of face boxes into 'generate_hdf5.py' or there is some mistake?

Thank you