Continual Adaptation of Visual Representations via Domain Randomization and Meta-learning

By Riccardo Volpi, Diane Larlus, Grégory Rogez. 2020

91 citations recorded by the source corpus.

Most standard learning approaches lead to fragile models which are prone to drift when sequentially trained on samples of a different nature—the well-known catastrophic forgetting issue. In particular, when a model consecutively learns from different visual domains, it tends to forget the past domains in favor of the most recent ones. In this context, we show that one way to learn models that are inherently more robust against forgetting is domain randomization—for vision tasks, randomizing the current domain’s distribution with heavy image manipulations. Building on this result, we devise a meta-learning strategy where a regularizer explicitly penalizes any loss associated with transferring the model from the current domain to different "auxiliary" meta-domains, while also easing adaptation to them. Such meta-domains are also generated through randomized image manipulations. We empirically demonstrate in a variety of experiments—spanning from classification to semantic segmentation—that our approach results in models that are less prone to catastrophic forgetting when transferred to new domains.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

General Incremental Learning with Domain-aware Categorical Representations (2022, methodology): “Furthermore, Meta-DR performs well on the iDomainNet and iDigits but does not perform well on iCIFAR-20.”
On Generalizing Beyond Domains in Cross-Domain Continual Learning (2022, neutral): “[35] which proposes continual domain adaptation, i.”
Recent Advances of Continual Learning in Computer Vision: An Overview (2021, neutral): “[38] further proposed a meta-learning strategy to consecutively learn from different visual domains.”
Domain Generalization: A Survey (2021, background): “For example, in biomedical applications where domain shift occurs between different patients’ data, it is impractical to collect each new patient’s data in advance [19]; in traffic scene semantic segmentation it is infeasible to collect data capturing all different scenes and under all possible weat”
Balancing Between Forgetting and Acquisition in Incremental Subpopulation Learning (2022, methodology): “The data augmentation comprises random resize crop, random horizontal flip, lighting, color jitter, etc. Note that such a heavy strategy is the same as the domain randomization (DR) method used in continual domain adaptation (CDA) [33] to achieve SOTA results.”
Continual Test-Time Domain Adaptation (2022, background): “[59] aims to continually adapt the unseen visual domain while alleviate the forgetting on the seen domain without retaining the source training data.”