Informative Image Regions Research Articles

Deep Semantic-Visual Alignment for zero-shot remote sensing image scene classification

Deep neural networks have achieved promising progress in remote sensing (RS) image classification, for which the training process requires abundant samples for each class. However, it is time-consuming and unrealistic to annotate labels for each RS category, given the fact that the RS target database is increasing dynamically. Zero-shot learning (ZSL) allows for identifying novel classes that are not seen during training, which provides a promising solution for the aforementioned problem. However, previous ZSL models mainly depend on manually-labeled attributes or word embeddings extracted from language models to transfer knowledge from seen classes to novel classes. Those class embeddings may not be visually detectable and the annotation process is time-consuming and labor-intensive. Besides, pioneer ZSL models use convolutional neural networks pre-trained on ImageNet, which focus on the main objects appearing in each image, neglecting the background context that also matters in RS scene classification. To address the above problems, we propose to collect visually detectable attributes automatically. We predict attributes for each class by depicting the semantic-visual similarity between attributes and images. In this way, the attribute annotation process is accomplished by machine instead of human as in other methods. Moreover, we propose a Deep Semantic-Visual Alignment (DSVA) that take advantage of the self-attention mechanism in the transformer to associate local image regions together, integrating the background context information for prediction. The DSVA model further utilizes the attribute attention maps to focus on the informative image regions that are essential for knowledge transfer in ZSL, and maps the visual images into attribute space to perform ZSL classification. With extensive experiments, we show that our model outperforms other state-of-the-art models by a large margin on a challenging large-scale RS scene classification benchmark. Moreover, we qualitatively verify that the attributes annotated by our network are both class discriminative and semantic related, which benefits the zero-shot knowledge transfer.

Exploring Spatial Diversity for Region-Based Active Learning.

State-of-the-art methods for semantic segmentation are based on deep neural networks trained on large-scale labeled datasets. Acquiring such datasets would incur large annotation costs, especially for dense pixel-level prediction tasks like semantic segmentation. We consider region-based active learning as a strategy to reduce annotation costs while maintaining high performance. In this setting, batches of informative image regions instead of entire images are selected for labeling. Importantly, we propose that enforcing local spatial diversity is beneficial for active learning in this case, and to incorporate spatial diversity along with the traditional active selection criterion, e.g., data sample uncertainty, in a unified optimization framework for region-based active learning. We apply this framework to the Cityscapes and PASCAL VOC datasets and demonstrate that the inclusion of spatial diversity effectively improves the performance of uncertainty-based and feature diversity-based active learning methods. Our framework achieves 95% performance of fully supervised methods with only 5 - 9% of the labeled pixels, outperforming all state-of-the-art region-based active learning methods for semantic segmentation.

Deep Neural Network-Based Cooperative Visual Tracking Through Multiple Micro Aerial Vehicles

Multicamera tracking of humans and animals in outdoor environments is a relevant and challenging problem. Our approach to it involves a team of cooperating microaerial vehicles (MAVs) with on-board cameras only. Deep neural networks (DNNs) often fail at detecting small-scale objects or those that are far away from the camera, which are typical characteristics of a scenario with aerial robots. Thus, the core problem addressed in this letter is how to achieve on-board, online, continuous, and accurate vision-based detections using DNNs for visual person tracking through MAVs. Our solution leverages cooperation among multiple MAVs and active selection of most informative regions of image. We demonstrate the efficiency of our approach through simulations with up to 16 robots and real-robot experiments involving two aerial robots tracking a person, while maintaining an active perception-driven formation. ROS-based source code is provided for the benefit of the community.

Modulation of microsaccades by spatial frequency during object categorization

The organization of visual processing into a coarse-to-fine information processing based on the spatial frequency properties of the input forms an important facet of the object recognition process. During visual object categorization tasks, microsaccades occur frequently. One potential functional role of these eye movements is to resolve high spatial frequency information. To assess this hypothesis, we examined the rate, amplitude and speed of microsaccades in an object categorization task in which participants viewed object and non-object images and classified them as showing either natural objects, man-made objects or non-objects. Images were presented unfiltered (broadband; BB) or filtered to contain only low (LSF) or high spatial frequency (HSF) information. This allowed us to examine whether microsaccades were modulated independently by the presence of a high-level feature – the presence of an object – and by low-level stimulus characteristics – spatial frequency. We found a bimodal distribution of saccades based on their amplitude, with a split between smaller and larger microsaccades at 0.4° of visual angle. The rate of larger saccades (⩾0.4°) was higher for objects than non-objects, and higher for objects with high spatial frequency content (HSF and BB objects) than for LSF objects. No effects were observed for smaller microsaccades (<0.4°). This is consistent with a role for larger microsaccades in resolving HSF information for object identification, and previous evidence that more microsaccades are directed towards informative image regions.

Highly Informative Natural Scene Regions Increase Microsaccade Production during Visual Scanning

Classical image statistics, such as contrast, entropy, and the correlation between central and nearby pixel intensities, are thought to guide ocular fixation targeting. However, these statistics are not necessarily task relevant and therefore do not provide a complete picture of the relationship between informativeness and ocular targeting. Moreover, it is not known whether either informativeness or classical image statistics affect microsaccade production; thus, the role of microsaccades in information acquisition is also unknown. The objective quantification of the informativeness of a scene region is a major challenge, because it can vary with both image features and the task of the viewer. Thus, previous definitions of informativeness suffered from subjectivity and inconsistency across studies. Here we developed an objective measure of informativeness based on fixation consistency across human observers, which accounts for both bottom-up and top-down influences in ocular targeting. We then analyzed fixations in more versus less informative image regions in relation to classical statistics. Observers generated more microsaccades on more informative than less informative image regions, and such regions also exhibited low redundancy in their classical statistics. Increased microsaccade production was not explained by increased fixation duration, suggesting that the visual system specifically uses microsaccades to heighten information acquisition from informative regions.