README.md

OPeN

Shiran Zada, Michal Irani

---

This is the unofficial implementation of OPeN in the paper Pure Noise to the Rescue of Insufficient Data: Improving Imbalanced Classification by Training on Random Noise Images (ICML 2022 spotlight) in pytorch.

No official implementation code yet exists

Code structures are similar with the code in https://github.com/kaidic/LDAM-DRW.git

How to run the code

1. Setting up the python environment

• PyTorch 1.2
• TensorboardX
• scikit-learn

2. Dataset

• Imbalanced CIFAR. The original data will be downloaded and converted by imbalancec_cifar.py.

3. Run the code on CIFAR-10/100 long-tail datasets

Use the following command to run the code on CIFAR-10/100 long-tail datasets. For comparison, other loss_type & train_rule(train methods) are prepared in the code.

• To train the ERM baseline on long-tailed imbalance with ratio of 100

python cifar_train.py --gpu 0 --imb_type exp --imb_factor 0.01 --loss_type CE --train_rule None --arch wide_resnet28_10

• To train the ERM Loss along with OPeN on long-tailed imbalance with ratio of 100

python cifar_train.py --gpu 0 --imb_type exp --imb_factor 0.01 --loss_type CE --train_rule OPeN --arch wide_resnet28_10

Results

	Baseline (ERM)	OPeN + AA
CIFAR-10 LT	79.6	85.25
CIFAR-100 LT	--	--

Reference

@article{DBLP:journals/corr/abs-2112-08810,
  author    = {Shiran Zada and
               Itay Benou and
               Michal Irani},
  title     = {Pure Noise to the Rescue of Insufficient Data: Improving Imbalanced
               Classification by Training on Random Noise Images},
  journal   = {CoRR},
  volume    = {abs/2112.08810},
  year      = {2021},
  url       = {https://arxiv.org/abs/2112.08810},
  eprinttype = {arXiv},
  eprint    = {2112.08810},
  timestamp = {Mon, 03 Jan 2022 15:45:35 +0100},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2112-08810.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}