train loss NaN !!

[9B8DY6]

I run train.py with your code but I ve got Nan in loss graph. I did not change dataset and any setting.

[YangShenglong-ai]

发现在莫一时刻，cx_gen loss 变成了nan，修改CX_style_loss.py 文件 def calc_relative_distances, 最后添加 relative_dist= torch.clamp(relative_dist, -1, 1, )虽然是强行限制，但应该不会报错了。希望作者可以给出更加有效的解决方法