Supervised Momentum Contrastive Learning for Few-Shot Classification

Abstract

Instance discrimination based contrastive learning has emerged as a leading approach for self-supervised learning of visual representations. Yet, its generalization to novel tasks remains elusive when compared to representations learned with supervision, especially in the few-shot setting. We demonstrate how one can incorporate supervision in the instance discrimination based contrastive self-supervised learning framework to learn representations that generalize better to novel tasks. We call our approach CIDS (Contrastive Instance Discrimination with Supervision). CIDS performs favorably compared to existing algorithms on popular few-shot benchmarks like Mini-ImageNet or Tiered-ImageNet. We also propose a novel model selection algorithm that can be used in conjunction with a universal embedding trained using CIDS to outperform state-of-the-art algorithms on the challenging Meta-Dataset benchmark.

The Best of Both Worlds: a Framework for Combining Degradation Prediction with High Performance Super-Resolution Networks

“…In the image classification domain, DNNs are known to be highly capable of learning invariant representations, enabling the construction of good classifiers [64]. However, it has been argued that DNNs are actually too eager to learn invariances [64]. This is because they often learn only the features necessary to discriminate between classes but then fail to generalise well to new unseen classes in a supervised setting, in what is known as "supervision collapse" [64,65].…”

Semantics-Guided Contrastive Network for Zero-Shot Object detection

“…For example, Khosla et al [52] directly used class labels to define similarity, where samples from the same class are positive and samples from different classes are negative samples. Majumder et al [53] devised few-shot learning with Instance discrimination based contrastive learning in a supervised setup. Inspired by the success of these methods, we first introduce contrastive learning mechanism to ZSD, and develop two contrastive learning subnets using high-level semantic information as additional supervision signals.…”

Distance-based Hyperspherical Classification for Multi-source Open-Set Domain Adaptation

“…The effectiveness of the contrastive self-supervised learned embeddings is generally evaluated by using the pretext feature model as starting point for a downstream supervised task. However, more direct ways to incorporate supervision are currently attracting large attention [17,52] and show how view invariance and semantic knowledge can be combined to get the best of both worlds in challenging scenarios as novelty detection [41], crossdomain generalization [61] or few-shot classification [27]. Current research is investigating ways to improve negative sampling [6], or present analyses to better understand the relation between contrastive learning and mutual information [53].…”