Why the result on the dataset CSS much is much less than the results in the paper?

Question

Why the result on the dataset CSS much is much less than the results in the paper?

Opened this issue 2 years ago · 2 comments

Hello, I would like to ask you 2 questions.
1、Why the result on the dataset CSS much is much less than the results in the paper?
Because of version iteration of pytorch, I make some modification to your code to run on the new version (1.10.2+cu102). The results on the fashion200 and MIT datasets are similar to those in the paper, but the result is much less on the CSS dataset.
2. The result on the CSS dataset is much better than results on the other two datasets. Why does the model perform so well on the CSS dataset?

Answer 1 · 2022-04-22T02:16:11.000Z

shchen0001 commented 2 years ago

Answer 2 · 2022-10-27T11:38:40.000Z

How long do you spend training？Can it be tested directly？I'm having trouble reproducing the code.Can you help me？