Tomohiro Mashita

Dr. Tomohiro Mashita

Associate Prof.
orcid

Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
13 June 2012 in Personal and Ubiquitous Computing
Verified
Journal Article
Closed access
PDF file
MIMayu Iwata
Mayu Iwata
HMHiroki Miyamoto
Hiroki Miyamoto
Takahiro HaraTakahiro Hara
Takahiro Hara
DKDaijiro Komaki
Daijiro Komaki
KSKentaro Shimatani
Kentaro Shimatani
HTHaruo Takemura
Haruo Takemura

13 November 2020 in Applied Sciences
Verified
Journal Article
Open access
PDF file
Tomohiro MashitaTomohiro Mashita
Tomohiro Mashita
TKTetsuya Kanayama
Tetsuya Kanayama
PRPhotchara Ratsamee
Photchara Ratsamee
Air conditioners enable a comfortable environment for people in a variety of scenarios. However, in the case of a room with multiple people, the specific comfort for a particular person is highly dependent on their clothes, metabolism, preference, and so on, and the ideal conditions for each person in a room can conflict with each other. An ideal way to resolve these kinds of conflicts is an intelligent air conditioning system that can independently control air temperature and flow at different areas in a room and then produce thermal comfort for multiple users, which we define as the personal preference of air flow and temperature. In this paper, we propose Personal Atmosphere, a machine learning based method to obtain parameters of air conditioners which generate non-uniform distributions of air temperature and flow in a room. In this method, two dimensional air-temperature and -flow distributions in a room are used as input to a machine learning model. These inputs can be considered a summary of each user’s preference. Then the model outputs a parameter set for air conditioners in a given room. We utilized ResNet-50 as the model and generated a data set of air temperature and flow distributions using computational fluid dynamics (CFD) software. We then conducted evaluations with two rooms that have two and four air conditioners under the ceiling. We then confirmed that the estimated parameters of the air conditioners can generate air temperature and flow distributions close to those required in simulation. We also evaluated the performance of a ResNet-50 with fine tuning. This result shows that its learning time is significantly decreased, but performance is also decreased.

1 October 2020 in IEICE Transactions on Information and Systems
Verified
Journal Article
Open access
PDF file
Tomohiro MashitaTomohiro Mashita
Tomohiro Mashita
KSKoichi Shintani
Koichi Shintani
KKKiyoshi Kiyokawa
Kiyoshi Kiyokawa
This paper introduces a user study regarding the effects of hand- and ocular-dominances to pointing gestures. The result of this study is applicable for designing new gesture interfaces which are close to a user's cognition, intuitive, and easy to use. The user study investigates the relationship between the participant's dominances and pointing gestures. Four participant groups—right-handed right-eye dominant, right-handed left-eye dominant, left-handed right-eye dominant and left-handed left-eye dominant—were prepared, and participants were asked to point at the targets on a screen by their left and right hands. The pointing errors among the different participant groups are calculated and compared. The result of this user study shows that using dominant eyes produces better results than using non-dominant eyes and the accuracy increases when the targets are located at the same side of dominant eye. Based on these interesting properties, a method to find the dominant eye for pointing gestures is proposed. This method can find the dominant eye of an individual with more than 90% accuracy.

1 December 2020 in IEEE Computer Graphics and Applications
Verified
Journal Article
Closed access
SSSota Shoman
Sota Shoman
Tomohiro MashitaTomohiro Mashita
Tomohiro Mashita
APAlexander Plopski
Alexander Plopski
PRPhotchara Ratsamee
Photchara Ratsamee
YUYuki Uranishi
Yuki Uranishi
Accurate camera localization is an essential part of tracking systems. However, localization results are greatly affected by illumination. Including data collected under various lighting conditions can improve the robustness of the localization algorithm to lighting variation. However, this is very tedious and time consuming. By using synthetic images, it is possible to easily accumulate a large variety of views under varying illumination and weather conditions. Despite continuously improving processing power and rendering algorithms, synthetic images do not perfectly match real images of the same scene, i.e., there exists a gap between real and synthetic images that also affects the accuracy of camera localization. To reduce the impact of this gap, we introduce “Real-to-Synthetic Feature Transform (REST)“. REST is a fully connected neural network that converts real features to their synthetic counterpart. The converted features can then be matched against the accumulated database for robust camera localization. Our experimental results show that REST improves matching accuracy by approximately 28% compared to a naiive method. This result guarantees a robust camera localization over various illuminations.

8 December 2021 in Robotics
Verified
Journal Article
Open access
PDF file
YTYusuke Takayama
Yusuke Takayama
PRPhotchara Ratsamee
Photchara Ratsamee
Tomohiro MashitaTomohiro Mashita
Tomohiro Mashita
Recently, several deep-learning based navigation methods have been achieved because of a high quality dataset collected from high-quality simulated environments. However, the cost of creating high-quality simulated environments is high. In this paper, we present a concept of the reduced simulation, which can serve as a simplified version of a simulated environment yet be efficient enough for training deep-learning based UAV collision avoidance approaches. Our approach deals with the reality gap between a reduced simulation dataset and real world dataset and can provide a clear guideline for reduced simulation design. Our experimental result confirmed that the reduction in visual features provided by textures and lighting does not affect operating performance with the user study. Moreover, by conducting collision detection experiments, we verified that our reduced simulation outperforms the conventional cost-effective simulations in adaptation capability with respect to realistic simulation and real-world scenario.

3 February 2026 in Sensors
Verified
Journal Article
Open access
PDF file
XSXuanmeng Sha
Xuanmeng Sha
Tomohiro MashitaTomohiro Mashita
Tomohiro Mashita
NCNaoya Chiba
Naoya Chiba
LZLiyun Zhang
Liyun Zhang
Self-supervised contrastive learning has demonstrated remarkable effectiveness in learning visual representations without labeled data, yet its application to 3D local feature learning from point clouds remains underexplored. Existing methods predominantly focus on complete object shapes, neglecting the critical challenge of recognizing partial observations commonly encountered in real-world 3D perception. We propose a momentum contrastive learning framework specifically designed to learn discriminative local features from randomly sampled point cloud regions. By adapting the MoCo architecture with PointNet++ as the feature backbone, our method treats local parts of point cloud as fundamental contrastive learning units, combined with carefully designed augmentation strategies including random dropout and translation. Experiments on ShapeNet demonstrate that our approach effectively learns transferable local features and the empirical observation that approximately 30% object local part represents a practical threshold for effective learning when simulating real-world occlusion scenarios, and achieves comparable downstream classification accuracy while reducing training time by 16%.