yonghaoxu/seanet

[AAAI 2019] Self-Ensembling Attention Networks: Addressing Domain Shift for Semantic Segmentation

seanet's Issues

Where website can read your paper?

Pretrained Model 404

I tried to open the URL of Pretrained Model, but get 404 for result.
Can you provide the fcn8s_from_caffe.pth, my email is [email protected], thank you very much!

您好！代码问题！

你好，请问训练的时候每轮结束所有类别的miou都是nan是因为什么呢

ablation experiments

尊敬的徐学长，你好！一个小小的问题。您肯定有做过attention机制的消融实验吧！额，请问下您的attention机制能带来多少的精度提升呢？

Teacher-Student model learning

Hi, Xu,

    I have read your paper and found it an interesting work. I have a question about your paper. In stardard self-ensemble learning, teacher model should be able to yield more accurate predictions to guide the student learning. How do you implement such mechenism in your paper?
    Thank you for your time!

Regards,
Yawei

关于训练的一些问题

在代码SEAN_GTA5.py 和 SEAN_Synthia.py中

    tgt_loader = data.DataLoader(
                    cityscapesDataSet(args.data_dir_target, args.data_list_target, max_iters=24966,                  
                    crop_size=input_size,
                    scale=False, mirror=False, mean=IMG_MEAN,
                    set='val'),
                    batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers,
                    pin_memory=True)

    val_loader = data.DataLoader(
                    cityscapesDataSet(args.data_dir_target, args.data_list_target, max_iters=None,                  
                    crop_size=input_size,
                    scale=False, mirror=False, mean=IMG_MEAN,
                    set='val'),
                    batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers,
                    pin_memory=True)

训练阶段和验证阶段都是使用Cityscape Val 数据集
我对此表示困惑

I do some ablation experiments!

Sorry to bother you again!!!

I do some ablation experiments for this paper. But I found some strange results.

Firstly, I run your original code for GTA5->Cityscapes. I got the following result:

Then, I remove your attention mechanism.

I got the following results:

Finally, after remove your attention mechanism, I remove your self-ensembling method.

I got the following results:

So I'm very confused about this. Can you give me some advice and opinion on this? Thanks so much!!!

the attention mechanism

You says "different regions in the images usually correspond to different levels of domain gap",I agree with you definitely!

Then you says "we introduce the attention mechanism into the proposed framework to generate attention-aware features".

After reading your paper and the code. I know you design an attention module in the segmentation network, and you use avgpool, UpsamplingBilinear2d, interpolation, aconv, sigmoid to build this module and get a mask, if the mask bigger than threshold 0.3(for example),then got 1,otherwise got 0.Then you obtain a M, You multiply M and consistency loss to selectively calculate the consistency loss.

But now I have a small question about the attention module. Why mask bigger than 0.3, then we focus on this pixel, and smaller than 0.3, then we ignore this pixel? Why your attention module can focus larger levels of domain gap and ignore smaller levels of domain gap? How do you make sure M can filter out the insignificant pixel？