Multi-level Adaptive Active Learning for Scene Classification

Abstract

Abstract. Semantic scene classification is a challenging problem in computer vision. In this paper, we present a novel multi-level active learning approach to reduce the human annotation effort for training robust scene classification models. Different from most existing active learning methods that can only query labels for selected instances at the target categorization level, i.e., the scene class level, our approach establishes a semantic framework that predicts scene labels based on a latent objectbased semantic representation of images, and is capable to query labels at two different levels, the target scene class level (abstractive high level) and the latent object class level (semantic middle level). Specifically, we develop an adaptive active learning strategy to perform multi-level label query, which maintains the default label query at the target scene class level, but switches to the latent object class level whenever an "unexpected" target class label is returned by the labeler. We conduct experiments on two standard scene classification datasets to investigate the efficacy of the proposed approach. Our empirical results show the proposed adaptive multi-level active learning approach can outperform both baseline active learning methods and a state-of-the-art multi-level active learning method.

Learning Loss for Active Learning

“…The entropy [47,31,19] of class posterior probabilities generalizes the former definitions. For SVMs, distances [52,53,27] to the decision boundaries can be used to define uncertainty. Another approach is the query-by-committee [49,34,18].…”

Active domain adaptation with noisy labels for multimedia analysis

“…Li and Guo [24] presented a novel multi-level AL approach to reduce the human annotation effort for training robust scene classification models. Different from most existing AL methods that can only query labels for selected instances at the class level, their approach established a semantic framework that predicted scene labels based on a latent object-based image representation, and was capable of querying labels at two different levels-the sceneclass level and the latent object-class level.…”

Temporal Coherence for Active Learning in Videos

“…A comprehensive survey of these frameworks and a detailed discussion can be found in [49]. Active learning has been successfully applied to a series of traditional computer vision tasks, such as image classification [28,24,14] (including medical image classification [46] and scene classification [35]), visual question answering (VQA) [37], image retrieval [62], remote sensing [8], action localization [19], and regression [11,25]. With a strong emphasis on image classification, active learning for object detection has received less attention than expected due to the difficulty to aggregate several object hypothesis at frame level.…”

“…Active learning has been mainly investigated for the im-age classification task [24,34,14,46,35,55,8]. Only few works have investigated active learning for object detection, even though the problem of active learning is more pertinent for object detection than for image classification since the labelling effort also includes the more expensive annotation of the bounding box [29].…”