airsplay/py-bottom-up-attention

Coordinates for feature prediction given boxes

eleonoragualdoni opened this issue · 1 comments

Feature prediction given boxes

If I want to predict features given box coordinates, following the script in your demos, are the coordinates to provide in the format XYXY, or in format XYWH?

Thanks!

The order is (x_left_corner, y_top_corner, x_right_corner, y_bottom_corner).

Here is a visualization. Suppose the box = [x0, y0, x1, y1], it annotates an RoI of:

0-------------------------------------
 |                                   |
 y0 box[1]   |-----------|           |
 |           |           |           |
 |           |  Object   |           |
 y1 box[3]   |-----------|           |
 |                                   |
H----------x0 box[0]-----x1 box[2]----
 0                                   W