jameszhou-gl / gpt-4v-distribution-shift

Code for ICLR'24 workshop ME-FoMo-How Well Does GPT-4V(ision) Adapt to Distribution Shifts? A Preliminary Investigation
https://arxiv.org/pdf/2312.07424.pdf
MIT License
33 stars 2 forks source link

有关于数据集 #8

Open yuu-Wang opened 3 months ago

yuu-Wang commented 3 months ago

您好,看见您的文章,gpt-4v-distribution-shift,非常感兴趣,特别想问一下

1.对于medical的四个数据集HAM10000,camelyon17,chest x ray14和xray他们应该属于什么类型的,目前知道camelyon17是domainbed里面的属于domain shift,ham10000属于in the wild吗,还有俩是什么呢。

2.对于这四个数据集划分的train,val,test的py文件有没有呢?

很急迫,望回复,很感谢,祝好!

jameszhou-gl commented 3 weeks ago
  1. not clear to understand your classification methods.
  2. We don't have train/val/test, since all datasets are used for testing. Checkout it as an example: https://github.com/jameszhou-gl/gpt-4v-distribution-shift/blob/3242ccbd7cb47de0dce147780b62ef9e89a938d7/data/process_medical.py#L16