Open OmerMagdy opened 2 years ago
In the code the implementation of the Giou loss is different than the original code:
w_intersect = torch.min(pred_left, target_left) + torch.min(pred_right, target_right) g_w_intersect = torch.max(pred_left, target_left) + torch.max( pred_right, target_right) h_intersect = torch.min(pred_bottom, target_bottom) + torch.min(pred_top, target_top) g_h_intersect = torch.max(pred_bottom, target_bottom) + torch.max(pred_top, target_top) ac_uion = g_w_intersect * g_h_intersect + 1e-7 area_intersect = w_intersect * h_intersect area_union = target_area + pred_area - area_intersect ious = (area_intersect + 1.0) / (area_union + 1.0) gious = ious - (ac_uion - area_union) / ac_uion
I think the fixed two lines should be :
g_w_intersect = torch.max(pred_right, target_right)-torch.min(pred_left, target_left) g_h_intersect = torch.max(pred_bottom, target_bottom) - torch.min(pred_top, target_top)
Can I get the confirmation of that fix ?
In the code the implementation of the Giou loss is different than the original code:
I think the fixed two lines should be :
Can I get the confirmation of that fix ?