Skip to content

Samyadeep/HardMD

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

HardMD++: Towards Understanding Few-Shot Performance on Difficult Tasks (ICLR 2023)

In this repository, we introduce FastDiffSel, an efficient algorithm to extract difficult few-shot tasks from Meta-Dataset and other large-scale vision datasets (e.g., ORBIT, CURE-OR, ObjectNet).

If you find our project helpful, please consider cite our paper:

@inproceedings{
basu2023hardmetadataset,
title={Hard-Meta-Dataset++: Towards Understanding Few-Shot Performance on Difficult Tasks},
author={Samyadeep Basu and Megan Stanley and John F Bronskill and Soheil Feizi and Daniela Massiceti},
booktitle={The Eleventh International Conference on Learning Representations },
year={2023},
url={https://openreview.net/forum?id=wq0luyH3m4}
}


To extract difficult few-shot tasks from Meta-Dataset Domains:


python fastdiffsel_md.py --dataset=meta_dataset --base_sources=aircraft --data-path /fs/cml-datasets/Meta-Dataset --arch dino_small_patch16 --fp16 --device cuda:0  --deploy weighted --weighted_step_size 0.2 --optimizer_epoch 1 --kmax 10 --query_opt 5 --joint_opt --sup 5  --md_sampling

Coming Soon!

data-loaders for extracted tasks from Meta-Dataset using FastDiffSel

Dependencies

pip install requirements.txt

To install the .h5 files in Meta-Dataset, follow the procedure in https://github.com/hushell/pmf_cvpr22

Correspondence to sbasu12@umd.edu

About

Algorithm to extract difficult few-shot tasks from Meta-Dataset and other large vision datasets

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages