I am very interested in the effectiveness of the baseline in the paper because the effect of Ostrack itself is not good (41.2, 53.6, 53.4). May I ask if your baseline has been modified based on Ostrack? The paper shows that your baseline was added with a convolutional layer after vit and before boxhead, but I am curious why such modifications can achieve such good results (even surpassing VIPT). I would like to know the specific operations of the author's baseline? Thank you very much for your answer.
I am very interested in the effectiveness of the baseline in the paper because the effect of Ostrack itself is not good (41.2, 53.6, 53.4). May I ask if your baseline has been modified based on Ostrack? The paper shows that your baseline was added with a convolutional layer after vit and before boxhead, but I am curious why such modifications can achieve such good results (even surpassing VIPT). I would like to know the specific operations of the author's baseline? Thank you very much for your answer.