search

WangWenhao0716 / TransHP

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

Question about the Linear layer dimension #2

Open TequilaDawn opened 3 months ago

TequilaDawn commented 3 months ago

Excuse me,could you tell me how the second parameter of the Linear layer represents how the dimension size is determined.😭 ![Uploading 请.png…]()

TequilaDawn commented 3 months ago
请
WangWenhao0716 commented 3 months ago

it is the number of coarse class at this hierarchy.

TequilaDawn commented 3 months ago

it is the number of coarse class at this hierarchy. i get it,thanks again

hongge831 commented 1 week ago

@WangWenhao0716 Hi I have the same question as @TequilaDawn , What is the philosophy behind setting the number of coarse categories at each layer? Could you please provide some indights for us ? Thanks you

TequilaDawn commented 1 week ago

@WangWenhao0716 Hi I have the same question as @TequilaDawn , What is the philosophy behind setting the number of coarse categories at each layer? Could you please provide some indights for us ? Thanks you This is determined by the nature of the ImageNet dataset, which divides the data into multiple levels, each with a different number of coarse categories.

WangWenhao0716 commented 1 week ago

@WangWenhao0716 Hi I have the same question as @TequilaDawn , What is the philosophy behind setting the number of coarse categories at each layer? Could you please provide some indights for us ? Thanks you

Hi, thanks for your interest. We do NOT set this manually. For each hierarchy dataset, the number of coarse and fine categories is pre-set, i.e., it is an attribute of the dataset. Hope that helps.