論文

In this paper, we address the problem of performing natural paste synthesis by color adjustment and image completion, in order to solve the completion problem that can specify an object appearing in a completion area. We propose a synthesis network that can extract the context features of the input image and reconstruct an image with the feature, making the inserted object appear in the completion region. In addition, we propose a ingenious method to make input images and learning method using Generative Adversarial Network (GAN) that do not require collection of high cost learning data. We show that color adjustment and image completion based on context features are executed at the same time, and natural pasting synthesis can be performed by using these proposal methods.

Sports video analysis enables professional teams to prepare and practice more efficiently than ever before. It allows the team to work on their weak points and analyze upcoming opponents' strategies. As segmentation is a well researched topic, we assume a sports video where each frame is segmented into actors. We show that the location of the actors is correlated to their action. We develop a model to predict each actor's action in a context consistent and location aware manner. The set of context actors is an unordered set, thus we propose a permutation invariant structure for our classifier. We evaluate a location and a context aware network for single image sports action recognition on two datasets. One is an existing dataset of volleyball games. We present a new dataset for actor recognition in rugby. We compare the results to using only the image of the actor.

In this paper, we propose a novel human-object interaction (HOI) representation for understanding plausible tool use; this is done by using an object-centric method. Everyday tools (such as kitchen or DIY utensils) have the ideal region and direction for being touched by hands and surrounding objects. By analyzing the human demonstrations, tactile histories that occurred on the tools are accumulated to the tool's surface as the "Tactile Log", which can help to achieve the plausible tool use by robot arms. We also proposed a dataset for the evaluation of generated tactile logs, which is called the Object-Centric Interaction (OCI) dataset consists of RGB-D videos of plausible tool use by human and 3D models of the tools used in the videos. We confirm that the precision of our tactile logging is 0.77.

We propose novel representations for functions of an object, namely Task-oriented Function, which is improved upon the idea of Afforadance in the field of Robotics Vision. We also propose a convolutional neural network to detect task-oriented functions. This network takes as input an operational task as well as an RGB image and assign each pixel an appropriate label for every task. Task-oriented funciton makes it possible to descibe various ways to use an object because the outputs from the network differ depending on operational tasks. We introduce a new dataset for task-oriented function detection, which contains about 1200 RGB images and 6000 pixel-level annotations assuming five tasks. Our proposed method reached 0.80 mean IOU in our dataset.

The latest generative adversarial networks (GANs) can generate realistic high resolution images. However, to the best of our knowledge, there are no GANs for generating 360-degree images. Therefore, this paper proposes the novel problem setting that by using a known area from the 360-degree image as an input, the remainder of the image can be completed with the GANs. To do so, we propose the approach of two-stage generation using network architecture with series-parallel dilated convolution layers. Moreover, we present how to rearrange images for data augmentation, simplify the problem, and make inputs for training the 2 stage generator. Our experiments show that these methods generate the distortion seen in 360-degree images in the outlines of buildings and roads, and their boundaries are clearer than those of baseline methods. Furthermore, we discuss and clarify the difficulty of our proposed problem. Our work is the first step towards GANs predicting an unseen area within a 360-degree space. nd

In this paper, we propose a method to estimate 6D pose (3D position and pose) of the everyday objects, especially Kitchen or DIY tools, even if there are not the same 3D model of the target object. We assume that the tools that belong to same category have common spatial relation in part-affordances, the role of each part. The proposed method uses this assumption for pose estimation. By minimizing two types of consistencies, the spatial relationship of part-affordances and the shapes, we have confirmed that the proposed method achieved higher performance compared with traditional pose estimation method.

The purpose of this study is to develop a tactics analysis system using image recognition for rugby. With the Rugby World Cup in 2019 and the Tokyo Olympics in 2020, demand for sports video analysis is increasing. Rugby has more complicated play such as dense play than other sports, and the ball is hidden between players, making it difficult to track. By developing a high-precision analysis technology for rugby with few research cases, we thought that it could be used for other sports and industrial fields other than sports. In this research, we propose a method that adds spatial information to time-series information as a new feature. Using the coordinates obtained by projectively transforming the match video onto the bird’s-eye view image, play classification was performed using the player position, the ball position, and the dense area position as feature amounts. Also, in order to further improve the detection accuracy of the boundaries between plays, attention was paid to the positional relationship of each player on the field.

The use of artificial intelligence (AI) as a tool supporting the diagnosis and treatment of spinal diseases is eagerly anticipated. In the field of diagnostic imaging, the possible application of AI includes diagnostic support for diseases requiring highly specialized expertise, such as trauma in children, scoliosis, symptomatic diseases, and spinal cord tumors. Moiré topography, which describes the 3-dimensional surface of the trunk with band patterns, has been used to screen students for scoliosis, but the interpretation of the band patterns can be ambiguous. Thus, we created a scoliosis screening system that estimates spinal alignment, the Cobb angle, and vertebral rotation from moiré images. In our system, a convolutional neural network (CNN) estimates the positions of 12 thoracic and 5 lumbar vertebrae, 17 spinous processes, and the vertebral rotation angle of each vertebra. We used this information to estimate the Cobb angle. The mean absolute error (MAE) of the estimated vertebral positions was 3.6 pixels (~5.4 mm) per person. T1 and L5 had smaller MAEs than the other levels. The MAE per person between the Cobb angle measured by doctors and the estimated Cobb angle was 3.42°. The MAE was 4.38° in normal spines, 3.13° in spines with a slight deformity, and 2.74° in spines with a mild to severe deformity. The MAE of the angle of vertebral rotation was 2.9° ± 1.4°, and was smaller when the deformity was milder. The proposed method of estimating the Cobb angle and AVR from moiré images using a CNN is expected to enhance the accuracy of scoliosis screening.

Person re-identification is a vital module of the tracking-by-detection framework for online multi-object tracking. Despite recent advances in multi-object tracking and person re-identification, inadequate attention was given to integrating these technologies to provide a robust multi-object tracker. In this work, we combine modern state-of-the-art re-identification models and modeling techniques on the basic tracking-by-detection framework and benchmark them on heavily occluded scenes to understand their effect. We hypothesize that temporal modeling for re-identification is crucial for training robust re-identification models for they are conditioned on sequences containing occlusions. Along with traditional image-based re-identification methods, we analyze temporal modeling methods used in video-based re-identification tasks. We also train re-identification models with different embedding methods, including triplet loss, and analyze their effect. We benchmark the re-identification models on the challenging MOT20 dataset containing crowded scenes with various occlusions. We provide a thorough assessment and investigation of the usage of modern re-identification modeling methods and prove that these methods are, in fact, effective for multi-object tracking. Compared to baseline methods, results show that these models can provide robust re-identification proved by improvements in the number of identity switching, MOTA, IDF1, and other metrics.