-
Notifications
You must be signed in to change notification settings - Fork 64
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
使用源码config训练CAN-DWAP,在CROHME14测试集上得到的ExpRate比论文中要低1.4% #27
Comments
随机种子与config内相同 为 20211024 |
我在测试中得到 52.23% ExpRate。可能是什么原因? |
尝试增加训练epoch数量 |
@lawsonxwl 太感谢了。我会尝试。 |
@lawsonxwl 我训练了240轮,代码没有任何改变,在2014 test 上只有54.67% ExpRate, 而且把counting loss 和 counting_preds 去掉之后,ExpRate 几乎不变,还是54%左右,和论文里的ablation study 完全对不上,不知道是哪里的问题,你有尝试过去掉counting loss 和 counting_preds 之后训练么 |
@eclickECNU 不,我到现在还没有尝试过。 |
大佬,能提供一下训练好的模型吗,我这边训练出来的效果达不到50%,不知道是不是batchsize设置小的问题。 |
作者是有自己的数据增强处理,但是这部分代码没有开源,所以达不到很正常 |
论文中无数据增强数据给的是57% 带上数据增强之后是65.58% |
是不是降低了batchsize啊,这个估计会有影响 |
使用的是论文的配置 batchsize为4 大了V100装不下 |
大佬你好 我看你在ICDAR2023的成绩很好 问下是使用的CAN模型吗? 有没有用啥trick 类似于beam search 和模型集成这种 |
config未作任何修改,Wordrate为0.9005,ExpRate为55.6%,问下大佬这个是啥情况?为什么复现不了呢
The text was updated successfully, but these errors were encountered: