argman/EAST

What does `geometry` contain?

Opened this issue · 0 comments

In run_demo_server.py L100, there is:

score, geometry = sess.run(
    [f_score, f_geometry], feed_dict={input_images: [im_resized[:, :, ::-1]]}
)

geometry is a (1, m, n, 5) array of floats which according to the paper should contain 4 distances from the pixel location to the top, right, bottom, left boundaries of the rectangle and the rotation angle respectively, I presume. I tried visualizing the geometry output however, it looks like something is different as by examining the resulting figure, I get noisy boxes not spread over text regions (which indicates that my top, right, bottom, left theory is incorrect). Can you please clarify what this geometry thing contains?