Abstract Long-tailed datasets are very frequently encountered in real-world use cases where few classes or categories (known as majority or head classes) have higher number of data samples compared to the other classes (known as minority or tail classes). Training deep neural networks on such datasets gives results biased towards the head classes. So far, researchers have come up with multiple weighted loss and data re-sampling techniques in efforts to reduce the bias. However, most of such techniques assume that the tail classes are always the most difficult classes to learn and therefore need more weightage or attention. Here, we argue that the assumption might not always hold true. Therefore, we propose a novel approach to dynamically measure the instantaneous difficulty of each class during the training phase of the model. Further, we use the difficulty measures of each class to design a novel weighted loss technique called ‘class-wise difficulty based weighted (CDB-W) loss’ and a novel data sampling technique called ‘class-wise difficulty based sampling (CDB-S)’. To verify the wide-scale usability of our CDB methods, we conducted extensive experiments on multiple tasks such as image classification, object detection, instance segmentation and video-action classification. Results verified that CDB-W loss and CDB-S could achieve state-of-the-art results on many class-imbalanced datasets such as ImageNet-LT, LVIS and EGTEA, that resemble real-world use cases.

Automatically describing video, or video captioning, has been widely studied in the multimedia field. This paper proposes a new task of sensor-augmented egocentric-video captioning, a newly constructed dataset for it called MMAC Captions, and a method for the newly proposed task that effectively utilizes multi-modal data of video and motion sensors, or inertial measurement units (IMUs). While conventional video captioning tasks have difficulty in dealing with detailed descriptions of human activities due to the limited view of a fixed camera, egocentric vision has greater potential to be used for generating the finer-grained descriptions of human activities on the basis of a much closer view. In addition, we utilize wearable-sensor data as auxiliary information to mitigate the inherent problems in egocentric vision: motion blur, self-occlusion, and out-of-camera-range activities. We propose a method for effectively utilizing the sensor data in combination with the video data on the basis of an attention mechanism that dynamically determines the modality that requires more attention, taking the contextual information into account. We compared the proposed sensor-fusion method with strong baselines on the MMAC Captions dataset and found that using sensor data as supplementary information to the egocentric-video data was beneficial, and that our proposed method outperformed the strong baselines, demonstrating the effectiveness of the proposed method.

Physiological signals such as heart rate can provide valuable information about an individuals state and activity. However, existing work on computer vision has not yet explored leveraging these signals to enhance egocentric video understanding. In this work, we propose a model for reasoning on multimodal data to jointly predict activities and energy expenditures. We use heart rate signals as privileged self-supervision to derive energy expenditure in a training stage. A multitask objective is used to jointly optimize the two tasks. Additionally, we introduce a dataset that contains 31 hours of egocentric video augmented with heart rate and acceleration signals. This study can lead to new applications such as a visual calorie counter.

A method for estimating the range between moving vehicles by using a monocular camera is proposed. Although most conventional methods use vertical triangulation, the proposed method uses both vertical and horizontal triangulation, which reduces errors due to vehicle's own pitching in the far distance. Unknown vehicle width is estimated by measuring three ranging parameters associated with an image captured by the camera, and the following distance is then computed by horizontal triangulation. Both vehicle width and following distance are dynamically updated during the vehicle-tracking process by single filtering. The proposed method runs in real time and can produce highly accurate estimation of following distance. The key contribution of this study is the coupled estimation of unknown vehicle width and following distance by sequential Bayesian estimation.

We propose a novel system for tracking multiple pedestrians in a crowded scene by exploiting single-row laser range scanners that measure distances of surrounding objects. A walking model is built to describe the periodicity of the movement of the feet in the spatial-temporal domain, and a mean-shift clustering technique in combination with spatial- temporal correlation analysis is applied to detect pedestrians. Based on the walking model, particle filter is employed to track multiple pedestrians. Compared with camera-based methods, our system provides a novel technique to track multiple pedestrians in a relatively large area. The experiments, in which over 300 pedestrians were tracked in 5 minutes, show the validity of the proposed system.

We propose a novel system for analyzing pedestrians' walking patterns by exploiting single-row laser range scanners that measure distances of surrounding objects by reflecting eye-safe laser beams. A walking model is built in the spatial-temporal domain to describe the periodicity of the movement of the feet, and a mean-shift technique is applied to recover model parameters. Compared with camera-based methods, our system provides a novel technique to analyze the behavior of pedestrians. The experiments show the validity of the algorithm.