Dongsheng Jiang

can you provide the parameter settings for finetune on 384 from 224 of image classification?

2

can you provide the parameter setting for finetune on 384 from 224?

unable to get the cifar10 accuracy

3

Hi, I just change your code with ResNet50(num_classes=10, resolution=(224, 224)), which end with a lower accuaracy of 90.15%. do you have other changes to get the 9511%?

have you ever try to train swin large?

3

I found swin_large_patch4_window7_224.yaml config file in your code. Here is an interesting question that how about the performance for larger mode?

使用github的代码和参数以及作者提供的数据，无法达到一模一样的动漫效果

5

没有任何改动，训练发现，每个iteration的效果，颜色都在变化，虽然也有动漫的效果，但是神韵和平滑性都不如作者提供的模型效果。还有什么trick吗？

knn eval of MAE

13

I eval vit_base of 500/1600 pretraining on imagenet1000 using knn metric. By loading all the pretained parameter with vit GAP method (not need cls token), the knn 20-NN result is...

it seems Unet in your code can outperform CE_Net in my training

2

it seems Unet in your code can outperform CE_Net in my training. I directly used your code and retrained the Unet, I got a accuracy of [acc: 0.956 | sen:...

我也在复现这篇文章

4

你好，这篇利用HDRNET来做样式迁移的应用，我也在复现。目前有一些效果。文章细节缺失，代码也没有公开。复现挺难的。从文章看，要用6次AdaIN，然后如果正则损失过小，也会产生很多偏色。所以regW我设置为100000。最后的效果其实更像是整体颜色的迁移，因为hdrnet本身就是保边滤波，无法迁移样式。

I want to know that your reimplments can have similiar results with the original paper? If ok, it is awesome!

when to release the training code of EfficientViT-SAM?

2

when to release the training code of EfficientViT-SAM?

it seems there are no train and test main code, please add them for reimplemention.