Official implementation of "Mining Multi-Label Samples from Single Positive Labels" (NeurIPS 2022). [arxiv]

We tested our algorithm in the following environment:

Anaconda
python 3.9.7
pytorch 1.10.1
numpy 1.20.3
torchvision 0.11.2

We prepare scripts for executing S2M sampling algorithm. To run the algorithm, you need to locate the checkpoints directory in the root directory of the project as follows:

┌── configs
├── scripts
├── src
└── checkpoints
    ├── GAN_cifar7to3
    ├── GAN_celeba7to3
    ├── cGAN_cifar7to3
    └── cGAN_celeba7to3

If you run the scripts below, sampling results on cifar7to3 and celeba7to3 will be saved in the results directory.

sh scripts/cgan_cifar7to3.sh
sh scripts/gan_cifar7to3.sh
sh scripts/cgan_celeba7to3.sh
sh scripts/gan_celeba7to3.sh