search

WangWenhao0716 / TransHP

[NeurIPS 2023] The official implementation of "TransHP: Image Classification with Hierarchical Prompting"
12 stars 0 forks source link

What do these twelve losses represent？ #1

Open TequilaDawn opened 6 months ago

TequilaDawn commented 6 months ago

When I try to train, I get a running error in the loss function, because the data dimension is incorrect, what do the twelve losses represent?Would you please answer it. BTW, I use the dataset CIFAR100 instead of image net1k, does that make a big difference？I would appreciate your help.

d91f414ca71b257d2ad799084bb6dfb0
WangWenhao0716 commented 6 months ago

Thanks for your interest. The 12 losses are for the 12 hierarchies of ImageNet. So, if you do not use ImageNet, it will make a big difference. And you should re-tune the weight and positions for prompting...unfortunately.

TequilaDawn commented 6 months ago

Thanks for your interest. The 12 losses are for the 12 hierarchies of ImageNet. So, if you do not use ImageNet, it will make a big difference. And you should re-tune the weight and positions for prompting...unfortunately.

Thanks for your interest. The 12 losses are for the 12 hierarchies of ImageNet. So, if you do not use ImageNet, it will make a big difference. And you should re-tune the weight and positions for prompting...unfortunately.

Thanks for your reply.Really help me a lot.