1. In the middle of training or at the beginning of saving CKPT, there will be the following problems

2. According to https://blog.csdn.net/qq_ 41046851/article/details/90552364_ No, it didn’t work

3. According to https://blog.csdn.net/jairana/article/details/83900226#41__ 294fine_ tune_ batch_ Normal = false, no success, learning_ Rate is already 0.0001

4. Exp/train_ on_ trainval_ After all the data in set/train are deleted (that is, the result of the last training is deleted), it runs successfully. Through the above changes, the problem of making mistakes after training for a period of time yesterday is solved. Now the training can run normally even after 1000 iterations