Question about dataset

[yangtiming]

Hello, I am interested in your HCT. And I try to understand your paper and code. Now, I have some problems about the code.

First, about pre-train: You know, in few shot learning, the most popular dataset is miniimagenet. Why do you train the HCT with imagenet? All of your codes run under the imagenet dataset rather than miniimagenet. [--data_path /path/to/imagenet/train]

Second, about main_hct_first.py Does main_hct_first.py run under the imagenet dataset? I think main_hct_first.py needs to run with miniimagenet dataset. But I try to run main_hct_first.py with miniimagenet dataset. However, the format of miniimagenet dataset is not suitable to main_hct_first.py.

I am looking forward to your reply.

[StomachCold]

Feel sorry to confuse you. Our repo is still being improved, so there may be some writing errors.

We did pretraining on the Few-shot learning dataset, but we had some writing errors editing the README. We do not use Imagenet as input, because Imagenet's training set contains test sets of mini_Imagenet and tiered_Imagenet. Testing mini directly with DINO's pre-trained model was 98% accurate because DINO was done on Imagenet. Finally, we changed mini_Imagenet from split to folder read, so you might have to do something with the data as well.

Thank you very much for your attention to our work, and I'm sorry to bring you trouble.

------------------ 原始邮件 ------------------ 发件人: "StomachCold/HCTransformers" @.>; 发送时间: 2022年4月12日(星期二) 晚上8:54 @.>; @.***>; 主题: [StomachCold/HCTransformers] Question about dataset (Issue #1)

Hello, I am interested in your HCT. And I try to understand your paper and code. Now, I have some problems about the code.

First, about pre-train: You know, in few shot learning, the most popular dataset is miniimagenet. Why do you train the HCT with imagenet? All of your codes run under the imagenet dataset rather than miniimagenet. [--data_path /path/to/imagenet/train]

Second, about main_hct_first.py Does main_hct_first.py run under the imagenet dataset? I think main_hct_first.py needs to run with miniimagenet dataset. But I try to run main_hct_first.py with miniimagenet dataset. However, the format of miniimagenet dataset is not suitable to main_hct_first.py.

I am looking forward to your reply.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

[yangtiming]

Thank you for your reply. So, in README, you said that "Download links of datasets could be found in the meta-transfer-learning repo". How does the format of miniimagenet change to the format, which is suitable to your repo.