Created by: seanria

训练过程如下： total number 500 step: 1, mIoU: 0.00036064465285404087 step 0, loss: 2.283701, step_time_cost: 24.394 Model is saved to output load from: output total number 500 step: 1, mIoU: 0.00040592799008661094 step 1, loss: 2.281190, step_time_cost: 22.383 Model is saved to output load from: output total number 500 step: 1, mIoU: 0.004588067474388075 step 2, loss: 2.277308, step_time_cost: 22.860 Model is saved to output load from: output total number 500 step: 1, mIoU: 0.07668137479753885 step 3, loss: 2.260504, step_time_cost: 22.590 Model is saved to output load from: output total number 500 step: 1, mIoU: 0.12360501949317738 step 4, loss: 2.255224, step_time_cost: 22.360 Model is saved to output load from: output total number 500 step: 1, mIoU: 0.12360501949317738 step 5, loss: 2.239232, step_time_cost: 22.390 Model is saved to output load from: output total number 500 step: 1, mIoU: 0.12360501949317738 step 6, loss: 2.223259, step_time_cost: 22.641 Model is saved to output load from: output total number 500 step: 1, mIoU: 0.12360501949317738 step 7, loss: 2.205665, step_time_cost: 22.690 Model is saved to output load from: output total number 500 step: 1, mIoU: 0.12360501949317738 step 8, loss: 2.187018, step_time_cost: 22.279

使用deeplabv3+训练，改过num_classes，其余训练和验证的核心代码没有动，之前训练几万步 loss 降到很低后跑验证集发现 mIoU 还是很低，调试后发现每张图片的 mIoU 达到某个值过后就不再变了。如上面这张验证图片的 mIoU 达到 0.12360501949317738 后就再也不变了，请问是什么情况？