Loss differences wrt paper

[lukszamarcin]

I'm not sure whether it's a bug or just difference in variable naming
In loss.py, you define the width and height of the union box as:

w_union = torch.min(d3_gt, d3_pred) + torch.min(d4_gt, d4_pred)
h_union = torch.min(d1_gt, d1_pred) + torch.min(d2_gt, d2_pred)

while in original paper it is:
wi= min(d2_gt,d2_pred) + min(d4_gt,d4_pred)
hi= min(d1_gt,d1_pred) + min(d3_gt,d3_pred)

Could you please clarify?

[SakuraRiven]

The variable naming does not matter as long as the decode function restore_polys (in detect.py) is align with encode order.

[lukszamarcin]

I see now, thanks for the clarification!