fungtion / dann_py3 Goto Github PK

Why not use the step of optimizer.zero_grad(),

Is it have to? or use or not use will not affect too much?

我想做可以检测森林火灾烟雾的DAN模型，那我是不是需要更换源域数据集呀，更换源域数据集的话都需要改什么部分呀

您好，我想请问一下，为了使源域和目标域距离拉近，随时训练的加深，域分类器的loss（error）是不是应该越来越大呀，

As shown in the image above, the domain classifier loss is almost constant throughout the training process. I use a ViT as the feature extractor, a linear layer as label classifier, and a two-layer MLP as the domain classifier.

What are the possible causes? and what are the typical loss curves for domain classifier?

Thanks

Hello, in README.md, the environment is Pytorch 1.6 Python 3.8.5, but in Dockfile is Pytorch 1.0.1 and uses pytorch/pytorch:1.0.1-cuda10.0-cudnn7-runtime as base images. So which one is right and if we should use Pytorch 1.6, then base images should use pytorch/pytorch:1.6.0-cuda10.1-cudnn7-runtime, right?

model.py中第23行使用了dropout2d，但是在全连接层情况下不是应该使用dropout吗？这个时候应该也没有channel来dropout了吧？望解答。

您好，我想请问一下，在main中len_dataloader取得是min（len(dataloader_souce),len(dataloader_target)）,一般来说目标域的数据要少于源域，那这样取循环的话，每次epoch源域不是都没有输入所有的样本嘛。那这样训练的话不就只使用了源域样本的一部分嘛？

Mnist_m cannot be downloaded
Baidu cloud disk has no password

_, domain_output = my_net(input_data=t_img, alpha=alpha)
err_t_domain = loss_domain(domain_output, domain_label)
err = err_t_domain + err_s_domain + err_s_label

I found here you backward the loss of three part...
but I think the taget dataset's class is the most important..
why this model's loss is not care about the taget dataset's class label?..

Hello @fungtion ,

I have been working on a problem which used CNN and performs regression in the end with the ground truth images.
Now, I want to incorporate DANN into my algorithm (CNN) so that it can withstand the domain variation.
Any suggestion in direction will be helpful.
One more thing, what is the effect of p on the DANN during training. Can I keep it as a fixed value?

I have a training set with 50k source images and 1k target images. Is DANN a good approach for this use case? If not, what is your recommendation?

下载的dataset是128Mb的文件 里面没有mnist_m_train/test等等这些东西 只有一个mnist_m.tar的文件这是要怎么做

Expected more than 1 value per channel when training, got input size torch.Size([1, 100])
I just changed the batch from 128 to 8, because MY GPU is not enough, but there is a mistake above, I don't know why

hi fungtion!

`class ReverseLayerF(F):

@staticmethod
def forward(ctx, x, alpha):
    ctx.alpha = alpha
    return x.view_as(x)`

As the title says, why just return x, thanks for your concern