Dataset Condensation with Contrastive Signals

This repository is the official implementation of Dataset Condensation with Contrastive Signals (DCC), published as a conference paper at ICML 2022. The implementation is based on (https://github.com/VICO-UoE/DatasetCondensation).

Prerequisites

• pytorch (1.2.0)
• numpy (1.15.1)
• torchvision (0.4.0)
• scipy (1.1.0)

Training and evaluation

To train the DCC (or DSAC) model in the paper, run this command:

python main.py --ipc <1, 10, or 50> --model ConvNet --dataset <CIFAR10, CIFAR100, or imagenet> (--imagenet_group <fine-grained dataset>) --method <DC or DSA> --contrast --save_path <save path name>

Please download ImageNet32x32 at (https://image-net.org/download-images)