CVDec 12, 2017

Data Distillation: Towards Omni-Supervised Learning

arXiv:1712.04440v1445 citations
Originality Incremental advance
AI Analysis

This addresses the challenge of leveraging large-scale unlabeled data for visual recognition, offering potential improvements over fully supervised methods, though it appears incremental as it builds on classic self-training ideas.

The paper tackles the problem of omni-supervised learning by proposing data distillation, a method that ensembles predictions from unlabeled data to generate training annotations, resulting in state-of-the-art models for human keypoint and object detection that surpass performance using only labeled COCO data.

We investigate omni-supervised learning, a special regime of semi-supervised learning in which the learner exploits all available labeled data plus internet-scale sources of unlabeled data. Omni-supervised learning is lower-bounded by performance on existing labeled datasets, offering the potential to surpass state-of-the-art fully supervised methods. To exploit the omni-supervised setting, we propose data distillation, a method that ensembles predictions from multiple transformations of unlabeled data, using a single model, to automatically generate new training annotations. We argue that visual recognition models have recently become accurate enough that it is now possible to apply classic ideas about self-training to challenging real-world data. Our experimental results show that in the cases of human keypoint detection and general object detection, state-of-the-art models trained with data distillation surpass the performance of using labeled data from the COCO dataset alone.

Code Implementations4 repos
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes