Active learning in CV图像分类是计算机视觉、模式识别领域的研究热点,在智能交通、安全监控、机器人导航等领域有着广泛的应用。在图像分类中,需要大量有标记的样本来训练稳定的分类模型,以实现对未知图像的准确分类。但是在实际应用中,有标记的图像数量非常之少,无标记的图像却随处可见,且图像的人工标记是件费时费力的工作。为了减少人工标记工作量,主动学习(Active Learning)技术被引入到图像分类中。主动学习的主要思想是：在大量未标记的样本中,采用某种策略,挑选少量最有信息量且最具代表性的样本交给专家进行标记。使用标记过的样本训练模型,实现对未知样本的准确分类。主动学习的核心技术是如何设计准则来挑选最具信息量的样本,以最大程度提升分类模型的性能。
The experiments on image classification and segmentation demonstrate that our model outperforms previous state-of-the-art methods and the initially sampling algorithm significantly improve the performance of our model
Our focus is on semantic segmentation; we believe that this work provides a highly promising research avenue towards other tasks in computer vision, including instance segmentation, object detection, activity understanding, or even visual-language embeddings
Through additional experiments we find that the difference in active learning performance can be explained by a combination of decreased model capacity and lower diversity of Monte Carlo Dropout ensembles
In this paper we introduced an approach to exploiting the geometric priors inherent to images to increase the effectiveness of Active Learning for segmentation purposes
Occurrences of facial actions are sparse within this data, our active learning approach has allowed us to acquire hand-labeled positive examples from many different individuals up to 20x faster
We propose perplexity based graph construction and a new hierarchical subquery evaluation algorithm to combat this variability, and to release the potential of Expected Error Reduction
We presented a new adaptive active learning approach which combines an information density measure with a most uncertainty measure together in an adaptive way to conduct instance selection
We have presented a discriminative probabilistic framework based on Gaussian Process priors and the Pyramid Match Kernel, and shown its utility for visual category recognition
