-
Notifications
You must be signed in to change notification settings - Fork 152
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Accurancy in dataset ic13_1015? #52
Comments
您试试再跑两个epoch?我发现不同的机子上测试似乎有点出入。 |
我稍微调研了一下其他开源的识别代码,也做了测试,例如CRNN和ASTER,似乎MORAN的开源代码和模型是最接近汇报的结果的:joy: |
再跑两个epoch会涨一点点,到92.3%。确实也非常接近您公布的结果了,算下来也就是差了10个样本。不过我用您提供的那个模型(demo.pth)在ic13_1015上测的结果怎么和您公布的不一致啊:joy: |
啊数据集的准确率是代码自动保存起来的。我随便挑了一个模型上传了,怕过拟合到这几个测试集上,遇到其他的样本泛化性能不好,被做实际应用的人嫌弃:joy: |
Soga, 可敬可敬:thumbsup:,学习了 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hi, what the training strategy(curriculum training or one-stage) was used for training the model you provided (demo.pth) ?
The accurancy of demo.pth in ic13_1015(LMDB file download from the link you provided) is 92.0%.
Meanwhile, I used the One-Stage training strategy and training with learning rate=1 for 3 epochs,then training with learning rate=0.1 for 1 epoch, all the synth data are used(NIPS2014 and CVPR2016), no other parameter was modified in train_MORAN.sh. The accurancy in ic13_1015 is also 92.0%.
How to imporve the performance of the model?(the accurancy can reach 93.2% as you mentioned) Any suggestion? Thanks a lot for your code.
The text was updated successfully, but these errors were encountered: