/AUTKC-optimization

Primary LanguagePython

Optimizing Partial Area Under the Top-k Curve: Theory and Practice

This is a Pytorch implementation of AUTKC optimization described in our paper:

Zitai Wang, Qianqian Xu, Zhiyong Yang, Yuan He, Xiaochun Cao and Qingming Huang. Optimizing Partial Area Under the Top-k Curve: Theory and Practice. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 45(4): 5053–5069, 2023.

And its Arxiv version can be found here.

The original version of our codes involves a non-standard IO operation, which is described later. Most recently, we provide a new demo that follows the standard learning pipline, which can be found in our library.

Dependencies

Please see the requirements.yml in the root folder.

Data and Train

We conduct the experiments on Cifar-10, Cifar-100, Tiny-imagenet-200, and Places-365. The method MemoryDataset.__init__() in the script dataset.py describes how to organize these datasets. What's more, to enable parallel computation, we create a shared memory to store the data. To this end, one should first run dataset.py before training or testing the model:

python dataset.py

Note that one should not kill the dataset process. Then, one can train the model via:

python main.py --dataset "cifar-100" --loss atop --weight_scheme Exp --resume checkpoints/***

Citation


@article{autkc,
  author    = {Zitai Wang and
               Qianqian Xu and
               Zhiyong Yang and
               Yuan He and
               Xiaochun Cao and
               Qingming Huang},
  title     = {Optimizing Partial Area Under the Top-k Curve: Theory and Practice},
  journal   = {{IEEE} Transactions on Pattern Analysis and Machine Intelligence},
  volume    = {45},
  number    = {4},
  year      = {2023},
  pages     = {5053--5069}
}