Squeeze, Recover and Relabel: Dataset Condensation at ImageNet Scale From A New Perspective
Document Type
Conference Proceeding
Publication Title
Advances in Neural Information Processing Systems
Abstract
We present a new dataset condensation framework termed Squeeze (Equation presented), Recover (Equation presented) and Relabel (Equation presented) (SRe2L) that decouples the bilevel optimization of model and synthetic data during training, to handle varying scales of datasets, model architectures and image resolutions for efficient dataset condensation. The proposed method demonstrates flexibility across diverse dataset scales and exhibits multiple advantages in terms of arbitrary resolutions of synthesized images, low training cost and memory consumption with high-resolution synthesis, and the ability to scale up to arbitrary evaluation network architectures. Extensive experiments are conducted on Tiny-ImageNet and full ImageNet-1K datasets1. Under 50 IPC, our approach achieves the highest 42.5% and 60.8% validation accuracy on Tiny-ImageNet and ImageNet-1K, outperforming all previous state-of-the-art methods by margins of 14.5% and 32.9%, respectively. Our approach also surpasses MTT [1] in terms of speed by approximately 52× (ConvNet-4) and 16× (ResNet-18) faster with less memory consumption of 11.6× and 6.4× during data synthesis. Our code and condensed datasets of 50, 200 IPC with 4K recovery budget are available at link.
Publication Date
1-1-2023
Recommended Citation
Z. Yin et al., "Squeeze, Recover and Relabel: Dataset Condensation at ImageNet Scale From A New Perspective," Advances in Neural Information Processing Systems, vol. 36, Jan 2023.