LAVA: Label-efficient Visual Learning and Adaptation
This work addresses the problem of label-efficient adaptation for visual learning across domains, which is incremental by building on recent innovations to improve performance in semi-supervised and few-shot settings.
The paper tackles multi-domain visual transfer learning with limited data by introducing LAVA, a method that combines self-supervised representation learning and pseudo-labeling with multi-crop augmentations, achieving state-of-the-art results on ImageNet semi-supervised protocol and 7 out of 10 datasets in multi-domain few-shot learning.
We present LAVA, a simple yet effective method for multi-domain visual transfer learning with limited data. LAVA builds on a few recent innovations to enable adapting to partially labelled datasets with class and domain shifts. First, LAVA learns self-supervised visual representations on the source dataset and ground them using class label semantics to overcome transfer collapse problems associated with supervised pretraining. Secondly, LAVA maximises the gains from unlabelled target data via a novel method which uses multi-crop augmentations to obtain highly robust pseudo-labels. By combining these ingredients, LAVA achieves a new state-of-the-art on ImageNet semi-supervised protocol, as well as on 7 out of 10 datasets in multi-domain few-shot learning on the Meta-dataset. Code and models are made available.