The exact clip adapter few shot acc for 11 datasets

[gordonhu608]

Hi authors, I'm a researcher at UCSD and I'm currently conducting the same research as you guys. I want to compare my model with the results from your model. Seems like there is no appendix or tables specifically showing the numbers of few shot (1,2,4,8,16) accuracy scores for 11 datasets. Is there a better way I can have these numbers than just trying to estimate them from your graphs?

[liyaowei-stu]

Me too. Hope authors provide more detailed results~ Moreover, do the authors have any plans to release the code for the t-SNE visualization.

[gaopengpjlab]

Sorry for the late response. We will update necessary information as soon as possible.

[ZrrSkywalker]

@gordonhu608 @liyaowei-stu The quantity results of CLIP-Adapter have been updated as here as CLIP-A.

[June01]

@gordonhu608 @liyaowei-stu The quantity results of CLIP-Adapter have been updated as here as CLIP-A.

Hi @gaopengcuhk, the zero-shot CLIP performance of eurosat in this log is eurosat: 37.52%, however, when I refer to the original paper of CLIP in Figure 8, it is around 60%. Would you like to clarify a bit? I got really confused.

[ZrrSkywalker]

@June01 Thanks for this question. We follow the code for data pre-processing in CoOp. Their reproduced zero-shot CLIP might have differences to the original CLIP paper.

[xuxiaoxxxx]

Hi, after I read the code, I find that this text encoder doesn't use the “residual ratio". The model architecture does not match the original paper, can you explain why？

@June01 Thanks for this question. We follow the code for data pre-processing in CoOp. Their reproduced zero-shot CLIP might have differences to the original CLIP paper.