'assert (boxes1[:, 2:] >= boxes1[:, :2]).all()' 在训练时发生

Thanks for the amazing work. I have some questions when training with your code. The `assert (boxes1[:, 2:] >= boxes1[:, :2]).all()` statement in the `generalized_box_iou` function occurred. After reading the code, I found that boxes1 is the predicted bounding box from an MLP layer, and I think the above assertion may occur during early training, which may break the training. I wonder if there are any mechanisms that can ensure that this does not happen.