The stochastic diffusion is an optimization method to minimize the potential function in the MAP-based motion estimations. This paper applies the stochastic diffusion to the MAP-based simultaneous motion estimation and segmentation in the hierarchical structure. According to the experiments, the proposed method can extract multiple objects with decreasing memory and computational burden which is the main problems in the stochastic diffusion.

抄録全体を表示

A generative model is presented for two types of person image generation in this paper. First, this model is applied to pose-guided person image generation, i.e., converting the pose of a source person image to the target pose while preserving the texture of that source person image. Second, this model is also used for clothing-guided person image generation, i.e., changing the clothing texture of a source person image to the desired clothing texture. The core idea of the proposed model is to establish the multi-scale correspondence, which can effectively address the misalignment introduced by transferring pose, thereby preserving richer information on appearance. Specifically, the proposed model consists of two stages: 1) It first generates the target semantic map imposed on the target pose to provide more accurate guidance during the generation process. 2) After obtaining the multi-scale feature map by the encoder, the multi-scale correspondence is established, which is useful for a fine-grained generation. Experimental results show the proposed method is superior to state-of-the-art methods in pose-guided person image generation and show its effectiveness in clothing-guided person image generation.

抄録全体を表示

Protein-protein interaction plays a critical role in biological processes. The identification of interacting proteins by computational methods can provide new leads in functional studies of uncharacterized proteins without performing extensive experiments. We developed a database for the potentially interacting domain pairs (PID) extracted from a dataset of experimentally identified interacting protein pairs (DIP: database of interacting proteins) with InterPro, an integrated database of protein families, domains and functional sites. In developing protein interaction databases and predictive methods, sensitive statistical scoring systems is critical to provide a reliability index for accurate functional analysis of interaction networks. We present a statistical scoring system, named “PID matrix score” as a measure of the interaction probability (interactability) between domains. This system provided a valuable tool for functional prediction of unknown proteins. For the evaluation of PID matrix, cross validation was performed with subsets of DIP data. The prediction system gives about 50% sensitivity and more than 98% specificity, which implies that the information for interacting proteins pairs could be enriched about 30 fold with the PID matrix. Itis demonstrated that mapping of the genome-wide interaction network can be achieved by using the PID matrix.

抄録全体を表示

Designing robots with human characteristics is a great challenge to researcher to fulfill the needs of robotic technology in human occupied environment. This research aims to enable a smooth and natural cooperative transfer by a human hand and a robot manipulator mimicking the same task performed by two humans. Prior to the development of the human-robot system, investigation on characteristics which generate a smooth cooperative transfer in a human-human system is crucial. Perceiving certain part of an in transfer, acts as a medium for the exchange of 's motion information between subjects. We have investigated the effect of perceiving different parts of the in transfer to the cooperative motion smoothness (smoothness is quantitatively evaluated using Minimum Jerk Model). The result suggests that observing the center part of the experimental produced more frequent, a smooth and natural motion for task executed in leftward/rightward and upward/downward direction. Moreover, the center part was associated with less rotation during smooth task in these directions. However, there is no significant difference between End and Center case for the same task executed in forward/backward direction. We also considered the importance of having information of cooperative task initiation signal and a target position to the cooperative task smoothness. Regardless of which part of the being perceived, smoother task is frequently generated when both signals are available to both subjects. Although the effect of End and Center case may be significant in human-human system; in human-robot system, both cases should be further tested and evaluated.

抄録全体を表示

The paper describes a method of extraction of grasping region of objects for automatic modeling by observing human- interaction. Human grasps and manipulates objects in everyday life. Home service robots and environmental cameras in intelligent space may obtain models by observing such human actions. The proposed method extracts grasping region of by observing human- interaction such as grasping and manipulating objects from the first person point of view with head mount camera.

抄録全体を表示

In this report, we discuss a method of measurement of subsurface scattering related to visual textures. Visual texture how a surface of

looks, such as glossy or matte, is related to the characteristics of light scattered or reflected with respect to incoming light. In previous study, a method presents an intermediate visual texture between glossy and matte has been proposed by juxtaposing the main surfaces of diffuse reflection and specular reflection. However, this method is not enough to reproduce actual visual texture because subsurface scattering is also important factor to visual texture. In this study, in order to grasp the relation between visual texture and subsurface scattered light and use findings for reproduction of visual texture, we considered the method to measure subsurface scattered light. The experiment showed subsurface scattered light was observed on translucent , and almost not on transparent and opaque .

抄録全体を表示

Floating detection and tracking has always been an important task for ensuring marine traffic safety and other surveillance purposes. To enable on scene observation, NMRI has developed a hybrid observing system. The system consists of a Laser Camera, a Night-vision Camera and an Infra-red (IR) Camera. By processing the collected IR images, this study aims at developing a floating objects track estimating system to support observation and provide warnings. The system should be able to automatically collect IR camera images and detect floating objects from those images, transferring the positions to the sea surface and estimate the objects tracks from their positions on consecutive camera images.

抄録全体を表示

Weakly supervised

detection, where a detector is trained with only image-level annotations, is attracting more attention. As a method to obtain a well-performing detector, the detector and the instance labels are updated iteratively. In this study, for more efficient iterative updating, we focus on the instance labeling problem, a problem of which label should be annotated to each region based on the last localization result, and two instance labeling methods are proposed. First, to solve the problem that regions covering only some parts of the tend to be labeled as positive, we find regions covering the whole focusing on the context classification loss. Second, considering the situation where the other objects in the image can be labeled as negative, we impose a spatial restriction on regions labeled as negative. Using these methods, we obtain the best results on the PASCAL VOC 2007 and 2012 datasets.

抄録全体を表示

The present study examined whether the spin of a horizontally translating target in an animation movie modulates the magnitude of forward displacement of a remembered final position of the target. The rotation of an axis bar corresponding to the diameter of a circular target represented the target's spin. There were three spin conditions: forward, backward, and without spin. The observers had to manually localize the vanished position of the target without eye movements. Experiment 1 showed that forward displacement was larger in the forward spin condition than in the backward spin condition and also confirmed that this modulation of forward displacement by spin was not due to observers' eye movements. Experiment 2 demonstrated that the modulation of forward displacement was not observed when horizontal translational motion was removed from the stimuli, suggesting that the interaction between the target's spin and the horizontal translational motion is critical. These results indicate that implicit friction due to spin modulates forward displacement without the involvement of eye movements.

抄録全体を表示

In order to be effective, it is essential for service robots to be able to recognize objects in complex environments. However, it is difficult for them to recognize objects autonomously without any mistakes in a real-world environment. Thus, in response to this challenge we conceived of an recognition system that would utilize information about target objects acquired from the user through simple interaction. In this paper, we propose an interactive recognition system using multiple attribute information (color, shape, and material), and introduce a robot using this system. Experimental results confirmed that the robot could indeed recognize objects by utilizing multiple attribute information obtained through interaction with the user.

抄録全体を表示

Convolutional neural networks (CNN) is driving advances in

recognition. In recent years, a general trend of CNN are built deeper and wilder that causes the networks being slow and high computational cost problem. Although the CNN leads high accuracy, it requires resource beyond an embedded system that has fixed resource. We propose a fine-tune network on embedded system for tactile recognition. Experiments on tactile image using a random-dot type optical tactile sensor.

抄録全体を表示

Recently, recognition is improved accuracy by Convolutional Neural Network(CNN) and growing recognition demand for automatic driving system and security and so on. However, one of the problem of recognition is to extract region which have various size, scale and form. This paper propose moving proposal for recognition. Using region grouping and scoring with optical flow, we can propose proposal for moving objects. The proposal experiments show effective results compared with previous method on the UCF crowd dataset.

抄録全体を表示

The paper describes a method for automatic acquisition of

appearances and grasping regions of everyday objects by observing human- interaction from first person point of view. Human grasps and manipulates objects in everyday life. Home service robots and robotic environmental systems may extract and obtain appearances by observing such human activity. However, the resolution of the acquired appearances may be low for recognition. Our proposed method obtain appearances and grasping regions by observing human- interaction using first person point of view with head mount camera. The acquired appearances are integrated by matching feature points and used for recognition.

抄録全体を表示

This paper considers how consumers pay attention to and participate in sales promotions. To achieve this purpose, we conducted experiments on sampling jam at a supermarket. We did comparison experiments with four conditions using an eye-catching

to encourage shoppers to notice the samples and a voting-style gimmick to encourage them to try the samples. Experimental results show that the and the gimmick were useful for encouraging shoppers to participate in the sampling. This tendency was noted among people of certain age groups, in pairs, and coming from the entrance.

抄録全体を表示

Edge devices with strict safety and reliability requirements, such as autonomous driving cars, industrial robots, and drones, necessitate software verification on such devices before operation. The human cost and time required for this analysis constitute a barrier in the cycle of software development and updating. In particular, the final verification at the edge device should at least strictly confirm that the updated software is not degraded from the current it. Since the edge device does not have the correct data, it is necessary for a human to judge whether the difference between the updated software and the operating it is due to degradation or improvement. Therefore, this verification is very costly. This paper proposes a novel automated method for efficient verification on edge devices of an

detection AI, which has found practical use in various applications. In the proposed method, a target existence detector (TOED) (a simple binary classifier) judges whether an in the recognition target class exists in the region of a prediction difference between the AI's operating and updated versions. Using the results of this TOED judgement and the predicted difference, an automated verification system for the updated AI was constructed. TOED was designed as a simple binary classifier with four convolutional layers, and the accuracy of existence judgment was evaluated for the difference between the predictions of the YOLOv5 L and X models using the Cityscapes dataset. The results showed judgement with more than 99.5% accuracy and 8.6% over detection, thus indicating that a verification system adopting this method would be more efficient than simple analysis of the prediction differences.

抄録全体を表示

This paper presents a modeling method for household objects for fetching. Grasps for household objects are selected according to the intended use of the . It is necessary to consider the functionality of the for modeling. Our method, first divides the body of a household into one or more parts in terms of its handling function. The functional parts are connected to each other with a link which has a joint attribute according to the parts configuration. The method offers an intuitive model for fetching.

抄録全体を表示