the error does not converge

[liustu]

I am reproducing the paper of you,but I found the error of training does not converge.The test error and similarity did not reach your results,on the dataset of kitti,the training error is about 0.15 and cannot go down.As to the dataset of chair,the training error is about 0.1 and cannot go down.In this condition,the error of testing was poor.In this case could you give me some guidance?Thank you very much!