In the field of object detection, particularly in medical imaging, the scarcity of data often poses a significant challenge to model performance. To address this issue, this study proposes a semi-supervised learning approach based on a generative model. We begin by fine-tuning a pre-trained generative model using our dataset to better adapt the generative model to our specific data distribution. The fine-tuned generative model is then used to generate additional unlabeled data. These generated unlabeled data, combined with the original dataset, are employed in a semi-supervised training process. Experimental results demonstrate that our method significantly enhances the performance of the object detection model, especially in scenarios with limited labeled data, such as medical imaging. By incorporating the generated unlabeled training data into the semi-supervised framework, we observed a notable improvement in model accuracy. Specifically, our experiments showed an increase of up to $6.92 \%$ after adding the generated iamges. Moreover, it is foreseeable that incorporating a higher proportion of generated unlabeled data could lead to even more significant improvements in performance.

As society develops, the use of smartwatches is becoming more common, and so too is the demand for more complex functions, including text input. Despite it being known that hand posture affects text input performance, most of the previous research has overlooked the influence of hand posture on smartwatch text input performance. Most of the current input methods have employed language and spatial models to improve text input performance. However, there is still a lack of research on the effect of hand posture on the input method after adding these models. Therefore, this study focused on the effect of the language model and spatial model when different hand postures (index finger support, no finger support, thumb and middle finger support, and thumb finger support) and different movement situations (walking and standing) were used.

Feedback latency during the use of interactive displays is an issue currently being considered in the HCI field. Several studies have focused on reducing latency using various approaches. This paper proposes a framework that uses a convolutional neural network to predict user touch points for interactive public displays. The framework predicts user touch events before the finger reaches the display surface to reduce the latency in feedback. As a training dataset, 1,651 tapping actions were collected from 18 participants in front of a display. The training of the convolutional neural network architecture was performed using the collected tapping actions. Validation test results showed that reasonable accuracy could be achieved at 390 ms before touching the display.

Most of the previous studies on smartwatch input methods were conducted in a stable experimental environment (while the user was sitting and using comfortable hand postures), and there have been few studies on the characteristics of input methods using different hand postures and in different body movements. This study evaluated smartwatch performance with different hand postures and in different body movements and compared the result with the ZoomBoard performance. We collected 16 participants' smartwatch input data and analyzed their tapping behavior with four hand postures (index finger support [IS], no finger support, thumb and middle finger support [TMS], and thumb support) and in two different body movements (walking and standing). We studied the input accuracy and speed with different hand postures and the distribution of the touchpoints. The results indicated that IS has the lowest error rate and fastest performance when the user is standing, and TMS when the user is walking. Also, the different hand postures had different touchpoint distributions. Therefore, this paper provides design considerations for future research on human-computer interaction on ultra-small screens, especially smartwatches.

Worldwide, 79.6 million people experience glaucoma, which can cause visual field loss to the individuals. Visual field examination plays an important role in the detection of glaucoma. However, current visual field examination approaches have disadvantages, such as requirements for expensive equipment, long testing time, location restrictions, and more. In the present study, we propose an assessment based on saccadic reaction time (SRT) to overcome the issues in the existing approaches. To confirm the effectiveness of our method, we simulated different stages of glaucoma using a visual field defect simulation system. The result of the visual experiment showed that the discrimination method using SRT can distinguish different symptoms with less testing time.

Visual field defects (VFD) can be caused by a variety of conditions. Checking and tracking the progression of VFD is an important part of an eye assessment. Although the use of standard automatic perimetry (SAP) is very popular for VFD diagnosis, it limits the population because of its high requirement for patients. We used a video-watching task as a replacement modality, which precludes the long period of fixation and uses the on-screen gaze to replace the button response. We developed a simulation system to mimic the different types of VFD in people with a normal pattern.We hypothesize that patients with VFD need more eye movement to compensate for the unseen area. We proposed a metric that indicates the gross eye movements toward a specific direction and found a significant difference between the VFD and normal pattern. Furthermore, we found videos that show the unique eye movement pattern in different eye conditions.

Small wearable information devices such as smart-watches are available everywhere. A user can input text in these devices while doing other tasks. However, eyes-free typing, that is, without looking at the input space, on these devices remains difficult to execute. This paper describes our project on developing eyes-free text input systems. Not only an auditory feedback feature was developed for the EdgeWrite text input system for round-face smartwatches, but also a simple template-based gesture recognition engine was implemented. The experimental results show that the developed system maintains reasonable text input speed and error rate for eyes-free under standing and walking conditions.

An origami tessellation display is an origami-based input/output device that has a flexible and unique deformable surface. This study achieves shape estimation and user gesture recognition for an origami-based surface without the use of external devices. Six tri-axis accelerometers are employed and embedded inside the surface, and shape estimation and gesture recognition models are constructed by means of neural networks and a random forest. Despite the small number of sensors and simple machine learning approaches, the models demonstrate sufficient estimation and recognition ability. In this paper, a framework in which shape estimation and gesture recognition are performed in real time is also presented.

This paper investigates smartwatch touch behaviors of users with different hand postures by conducting a laboratory experiment with 12 participants and analyzing their tap events on a smartwatch with four postures: the no-support (NS) posture, thumb support (TS) posture, index-finger-support (IS) posture, and thumb-and-middle-finger-support (TMS) posture. The participants' movement time and the error rate of the tapping tasks were measured for each posture and for two target positions (edge and center) and three target sizes (large, medium, and small). The results indicate much faster response times for the small, center targets for the TS and TMS postures compared with the NS posture. Based on this result, this paper proposes five design guidelines for arranging touch targets on a smartwatch touchscreen based on four postures.

This paper introduces a smartphone-based system that encourages photography by enhancing a user's awareness of the surrounding environment during a stroll in a city. When a user approaches a pre-registered location for photography, the system notifies the user by vibrating the smartphone. The system only displays the distance to the location and a photograph of the location on the smartphone screen; it does not provide detailed information, such as that given by a map. A user study performed during a stroll in a city confirmed the application of this system. The results demonstrate that ambiguous information provided by the system increased the opportunity for photography and fortuitous discovery.

This paper presents our project of designing Edge Write text entry methods for Japanese language. We are developing two versions of Edge Write text entry method for Japanese language: Roman-Kana Edge Write and Katakana Edge Write. Roman-Kana Edge Write is a straightforward application of the original Edge Write to Japanese phonograms. In Roman-Kana Edge Write, each Japanese phonogram is represented as a combination of a consonant and vowel, each of which is specified by an English character in the original Edge Write. Katakana Edge Write specifies the line stroke directions and writing order of the Japanese Katakana character. The ideal corner sequence pattern of Edge Write for each Katakana character is designed based on its line stroke directions and writing order.

The performances of both audio-visual content and systems are often evaluated by the sense of presence, which can be divided into two aspects: content presence and system presence. In our previous study, we constructed neural-network-based Kansei estimation models to evaluate content presence. Herein we aim to incorporate system presence into the Kansei models. We initially examined five audio reproduction methods, which simulate different systems, and conducted subjective evaluation experiments using 12 audio-visual content items. The experiments indicate the audio reproduction method influences both the audio-only and audio-visual conditions, but the effect is larger in the audio-only condition. Thus, we introduced four features related to the spatial impression of sound as new inputs into the previous Kansei estimation models. The expanded models successfully estimated both the content presence and system presence regardless of the condition. Hence, these models can quantitatively estimate the sense of presence in both audio-visual content and systems.

The term `presence' is widely used to express the performances of audio-visual content and systems. Until now, many advanced audio-visual reproduction systems have been proposed to enable the perception of a higher degree of presence. Methodologies for evaluating the sense of presence derived from these systems also play an important role in the further evolution of audio-visual content and systems. This paper describes the construction of Kansei models which quantitatively evaluate the degree of presence in audio-visual content. Subjective evaluation experiments were conducted to investigate the relationship between the features of audio-visual content and the perceived sense of presence using 40 different types of audio-visual content. On the basis of evaluation scores and extracted audio-visual features, models were separately constructed for audio, visual and audio-visual content using neural networks. The performance tests of the constructed models demonstrated sufficient accuracy in estimating the sense of presence and statistically significant correlations between the model output and the subjective evaluation scores.

Consideration of the preferences of multiple group members is a fundamental requirement for tour route planning. This paper describes the implementation of a tour route planning support system to simplify the planning process. The system recommends combinations of tourist attractions for a group considering the preferences of group members. First, a Kansei database linking tourist attractions and their characteristics is constructed based on the results of Kansei evaluation experiments. Next, the system is implemented using an evolutionary algorithm with the Kansei database. After the implementation, performance of the system is verified using several test samples. The results demonstrate that our system has sufficient ability to propose appropriate and wide variety of tourist attractions.

Townscape colours have been a main issue in urban-development. For townscape colours, keeping colour harmony within the environment is a common goal. Expressing characteristics and impressions of the town in townscape colours are other meaningful goals. This paper describes the colour planning support system intended to improve town-scapes. The system offers some colour combination proposals based on three elements: colour harmony, impressions of the townscape, and cost for the change of colours. First, we develop evaluation models to quantify colour harmony and impression of the townscape from the approach of Kansei engineering. Next, the system is constructed using an evolutionary algorithm and the two evaluation models. After the construction, performance tests are conducted. The results show that our system achieved sufficient ability to propose appropriate colour combinations with minimum colour changes.