Rosun

Hi, @quark0 I meet the same problem. The inplace parameter of each layer in my model is set to false. Could u give me some suggestion for solving this issue....

Hi, @xxsgcjwddsg , have u solved this issue?

the auxiliary loss should follow res4b_22

@blackcow: Thank you for your attention. Recently, we are busy with ECCV. We will release it as soon as possible.

@xiaoguoguo2018 关于问题1： label是指单通道的map, 范围是0,..., Num_Class -1. 关于问题2：在验证测试时label一般不做数据增强，一方面我们期待模型为输入图像做分割预测，输入你怎么变换都可以，但label要保持原始的。另一方面，如果label也做数据增强，那精度评估就不公平了，因为不同人使用数据增强也有差异。

@rock4you There may be some problems with the current training code, which we are trying to solve.