While data collection for model training is relatively easy, curating high-quality datasets from large collections is inherently challenging. Coreset estimation methods address this problem by identifying the best data subsets to train models on. However, existing approaches often require proxy model training, extensive annotations, or task-specific tailoring. Unsupervised alternatives typically exploit geometric properties in foundation model feature spaces, but are limited to pairwise image-to-image comparisons or comparisons against a fixed subset. We introduce CoreDNA, an unsupervised, task-agnostic coreset estimation approach that pools neuron activations from vision foundation models into per-image and per-dataset distributions. By comparing each image's distribution to the dataset-level distribution, CoreDNA can capture both low-level visual structure and high-level semantic information across layers. We find that high-quality samples concentrate in the tails of the last layer's activation statistics, which we validate by relating them to training-based coreset methods. Building on these findings, we develop a computationally efficient selection mechanism to construct task-agnostic, high-quality coresets for selective annotation and model training. Across image classification, semantic segmentation, depth estimation, and imitation learning for end-to-end driving and mobile manipulation, CoreDNA matches or outperforms the state of the art, demonstrating unprecedented generalization across perception and action tasks. These results highlight the effectiveness of compact activation-based representations for data selection.

Per-image pattern extraction. CoreDNA passes each unlabeled image through a frozen vision foundation model and pools the neuron activations of every layer into a compact histogram. Each image is thereby described by per-layer statistics: the mean, the standard deviation and the maximum of its activations. The same statistics pooled over the whole collection give a dataset-level distribution.

Selection. Comparing each image's distribution to the dataset-level distribution captures both low-level visual structure and high-level semantic information across layers. We find that high-quality samples concentrate in the tails of the last layer's activation statistics, which we validate by relating them to training-based coreset methods. We therefore select coresets in the space spanned by the last layer's mean and standard deviation and the cross-layer dispersion, either by minimizing a sliced Wasserstein distance to the dataset distribution or by farthest-point sampling. No labels, proxy model training or task-specific tuning is needed, so the same coreset pipeline serves perception and action tasks.

CoreDNA is evaluated on image classification (CIFAR10), semantic segmentation (Cityscapes, PascalVOC), monocular depth estimation (Cityscapes-DVPS, SemanticKITTI-DVPS), and imitation learning for end-to-end driving (NAVSIM) and manipulation (LIBERO-10). Across these tasks it matches or outperforms the state of the art. Please see the paper for full quantitative results and ablations.

Julia Hindel, Benjamin Ramtoula, Kyuhwan Yeon, Nick Heppert, Paul Newman, Daniele De Martini, and Abhinav Valada
Diversity Without Labels: Task-Agnostic Coreset Selection Across Perception and Action
arXiv preprint arXiv:2610.23642

@article{hindel26coredna,
			title={Diversity Without Labels: Task-Agnostic Coreset Selection Across Perception and Action},
			author={Julia Hindel and Benjamin Ramtoula and Kyuhwan Yeon and Nick Heppert and Paul Newman and Daniele De Martini and Abhinav Valada},
			journal={arXiv preprint arXiv:2610.23642},
			volume={arXiv:2610.23642},
			year={2026}
		  }