zhangchbin/onlinelabelsmoothing

The official code for the paper "Delving Deep into Label Smoothing", IEEE TIP 2021

onlinelabelsmoothing's Issues

相似工作：Label Embedding Network

您好！在读您的论文时，感觉和北大语言所孙栩老师2017年的工作非常像，附上文章名称和链接，供您参考。

论文名称和链接：
Label Embedding Network: Learning Label Representation for Soft Training of Deep Networks
http://arxiv.org/abs/1710.10393

When will the code come?

I have wrote a code from your paper, and tried to reimplement the results on ImageNet, but I can't get a good result. I just get the baseline via your training strategy. But one thing, I have not considered about synchronization in my code, and I use 4 Tesla V100 GPUs.
Waiting for your code.
Thanks.

论文中BYOT的实现问题

作者您好，

我看到这篇论文中使用了BYOT，我读了BYOT的原文，但是没有找到官方开源的代码，只在github看到一个实现:
https://github.com/luanyunteng/pytorch-be-your-own-teacher。

但是这个实现和原文中的结果差别挺大，您论文中BYOT用ResNet50在CIFAR-100的表现(80.8%)比原作者的还要好一些(80.56%)

想问一下您是怎么实现BYOT的？谢谢！

大佬，代码什么时候开源呢？想试试

类别不均衡的数据，或许有奇效

I have a question about noisy cifar experiment.

Hi, Chang-Bin Zhang.

I read your paper with great interest.
I think your method of generating soft labels considering category relationships is very excellent.
I would like to reproduce the noisy CIFAR100 experiment, is it possible for you to share the code?
I am very interested in your work and would be happy to provide it.

I would appreciate an answer.
Thank you in advance.

solver.py中清空cur_loss的疑问

您好，请问在solver.py中193行处，为什么要在这个时机要清空cur_loss，请指教
if (i+1) % (len(dataset)//2//self.args.batch_size) == 0:
print('%s [epoch %d/%d, iter %d/%d] lr = %f cur_loss = %f avg_loss = %f' % (time_now, epoch, self.args.epochs, i, len(dataloader), optimizer.param_groups[0]['lr'], cur_loss/100, loss_recoder.avg))
cur_loss = 0

tabular data/ noisy instances/ new datasets

Hi,
thanks for sharing your implementation. I have some questions about it:

1. Does it also work on tabular data?
2. Is the code tailored to the datasets used in the paper or can one apply it to any data?
3. Is it possible to identify the noisy instances (return the noisy IDs or the clean set)?

Thanks!

The batchsize in Object detection

Thanks for your great job! Can I ask the batch size in the implementation of object detection?