search

KMnP / vpt

❄️🔥 Visual Prompt Tuning [ECCV 2022] https://arxiv.org/abs/2203.12119
Other
1.04k stars 91 forks source link

Total tunable parameters #17

Closed 111chengxuyuan closed 2 years ago

111chengxuyuan commented 2 years ago

image Hello, I'd like to ask you a quesetion, what is the unit of the "Total params" in the experiment table? is it "M"? For example, the "Total params" of VPT-deep is 1.18x, does it mean 1.18M?

Charleshhy commented 2 years ago

You can refer to https://github.com/KMnP/vpt/issues/10. "Total params" is the number of params including backbone and the other trainable parameters for a total of 19 (VTAB) + 5 (FGVC) tasks.

KMnP commented 2 years ago

Thanks @Charleshhy! Let me know if you have other questions! @111chengxuyuan

KMnP commented 2 years ago

Close this issue for now. Feel free to re-open it if you have other questions!

ChengHan111 commented 1 year ago

Hi, I am wondering if there is a specific reason to include a backbone param within VPT total param calculation. I am getting a little bit confused by calculating with one backbone param + the other trainable parameters

Tsingularity commented 1 year ago

@ChengHan111 "Total Params" here refers to all the parameters we need to store for all the 24 downstream classification tasks. Since VPT doesn't modify the backbone params, we only need to store one copy of it. Full-finetuning will need to store 24 finetuned copies.

ChengHan111 commented 1 year ago

ation tasks. Since VPT doesn't modify the backbone params, we only need to store one copy of it. Full-finetuning

Thanks! Now I get it.