コンピュータマインド List of Products and Services

We developed an application that operates on ROS, integrating LiDAR-SLAM for mapping, self-localization, and underground data measurement devices. This application enables accurate understanding of the measurement information from underground data measurement devices and their positional relationship above ground, while also allowing for visual recording of the surrounding environment. Additionally, by using ROS (Robot Operating System), we have successfully facilitated smooth data transmission and reception, coordinate transformation, and sensor information output between various sensors and devices. 【Development Example】 ■OS: Ubuntu 16.04 LTS ■Development Period: 1 year ■Number of Developers: 3 ■Programming Languages: C++, Python ■Framework Used: ROS ■Related Technology: Point Cloud Matching *For more details, please refer to the PDF document or feel free to contact us.

• others

The treatment of eye diseases is advancing daily, but the information from biological tissues is not necessarily sufficient, and there are limitations in understanding the pathological conditions. In this development, we are collaborating with manufacturers of testing devices to obtain new tissue characteristic information, and we are using advanced technologies such as Deep Learning and GPGPU to make the most of that information, demonstrating means to understand pathological conditions from multiple perspectives and addressing this issue. In conventional methods, it took approximately one hour to calculate the data from a single examination, but we have successfully re-implemented this calculation on GPGPU, reducing the time to about five minutes. 【Development Example】 ■ OS: Windows 10 ■ Development Period: 5 years ■ Number of Developers: 3 ■ Development Languages: Qt, C++, Python ■ Networks Used: DeepLabV2, V3 *For more details, please refer to the PDF document or feel free to contact us.

• others

To put technologies such as object detection using Deep Learning into practical use, systematization is essential. At our company, in addition to Deep Learning, we have proposed software based on the hardware control and communication technologies we have cultivated over the years. Furthermore, we provide all the necessary elements for systematization in a one-stop manner, including the selection of hardware such as cameras suitable for the system. [Development Example] ■ OS: Windows, Linux ■ Development Period: Poc 3 months, Development 4 months (prototype) ■ Number of Developers: 3 people ■ Programming Languages: C#, C++ ■ Network Used: YOLO v4 tiny *For more details, please refer to the PDF materials or feel free to contact us.

• others

Our company has successfully detected a large number of foreign objects that were previously missed by applying Deep Learning object detection technology to images captured by inspection equipment. In image processing, it is common to handle complex processing steps sequentially, which can lead to longer processing times per inspection, ultimately affecting overall production efficiency. With this technology, we have achieved high-speed inspections of approximately 100ms or less per inspection (per image), allowing us to improve inspection accuracy without compromising production efficiency. [Development Example] - OS: Windows 10 Pro - Development Period: 6 months - Number of Developers: 2 - Programming Languages: C#, C++, Python - Networks Used: SSD, Yolov3 *For more details, please refer to the PDF document or feel free to contact us.

• others

Image recognition technology using Deep Learning is expected to be applied in visual inspection in industries such as manufacturing due to its high detection performance and generalization capabilities. Our company has achieved results in defect detection using object detection with Deep Learning through a Proof of Concept (PoC) in an automated visual inspection project for sanitary ware in collaboration with LIXIL. We are currently advancing towards trial operations for line implementation. Regarding this development initiative, we presented jointly with LIXIL at the interactive session of the 2020 Annual Conference of the Japanese Society for Artificial Intelligence. [Development Example] - OS: Windows 10 Pro - Development Period: 2018 onwards - Number of Developers: 1 for PoC / 2 for main development (planned) - Development Languages: C#, Python - Network Used: SSD512 (VGG16) *For more details, please refer to the PDF document or feel free to contact us.

• others

Many people are considering the application of AI technology to visual inspection (quality control). Here are the key points for utilizing AI in visual inspection. ◆ Benefits after implementation - Reduction of labor costs - Improvement of production efficiency - Standardization of judgment criteria - Increase in detection rates ◆ Challenges when implementing - Difficulty in setting imaging conditions - Difficulty in clarifying judgment criteria - Uncertainty about whether detection is possible with AI - Some AI vendors may charge regular fees for AI usage ◆ Solutions with DeepEye! - Achieving a low price of 980,000 yen - One-time purchase model, so the payment amount is fixed regardless of usage - A free rental period of two weeks is provided, allowing for the acquisition of evaluation materials.

• Other Software

I will introduce successful development when requesting an AI vendor. 1. Goals and processes are specific. Development with specific goal setting, such as "I want to automate the □ task of the △ line of the 〇〇 product," and clear processes often leads to success. 2. The data acquisition environment is well-established. The purity of data greatly influences development results. Therefore, when the data acquisition environment is well-established, the success rate can be said to be higher. 3. AI is viewed as part of the system. It is important to see AI as a means to solve needs and to consider the usage environment. I have introduced three points, but it can be difficult for customers to implement the above on their own. By using our "DeepEye," it is possible to easily engage in AI development. - It is a one-time purchase model, making budgeting easier. - Even if it is unclear whether AI can be applied, it can be verified multiple times. - Even if there are many issues, the cost remains the same. Additionally, our company also provides AI consulting, and we have a support system in place to help solve our customers' challenges.

• Other Software

In recent years, tools and services that allow for the easy creation of Deep Learning models have been made available. With this ease of access, there are likely already individuals who have begun developing Deep Learning models. The next challenge after creating a Deep Learning model is "improving accuracy." When the expected accuracy is not achieved, the following actions can lead to performance improvements: 1. Change various hyperparameters. 2. Increase the amount of training data. 3. Increase the "patterns" in the training data. For example, differences in lighting brightness, size, and speed. 4. Perform preprocessing to clarify the features of the data. 5. Compare and consider multiple types of models to use. 6. Review the class design of the model. 7. Build an ensemble model that combines multiple models. However, carrying out these tasks requires specialized knowledge, as well as time and effort to prepare the data. By using our "DeepEye," you can easily implement these accuracy improvement strategies.

• Other Software

An all-in-one package PC capable of image classification and object detection using Deep Learning technology. - No detailed knowledge of Deep Learning is required. - No preparation or knowledge of the Deep Learning operating environment (Linux) is necessary. - With a user-friendly GUI, you can easily execute a series of processes for creating training data, building learning models, and running inferences, allowing for repeated validation within your company.

• Other Software

We propose solutions to improve productivity through the introduction of "deep learning," one of the machine learning techniques. By creating a network that mimics the structure of the brain's neural transmission network, we can perform calculations from the input layer to the output layer. By adjusting the parameters used during the calculations, we can achieve our objectives. Additionally, even in cases where the model cannot be expressed in mathematical formulas, it can be represented through the network, allowing computers to make judgments on matters that could only be assessed by human senses until now. [Features] ■ If you provide data, a simple trial evaluation is possible. ■ The above mechanism can be utilized for budget allocation and internal approvals within your company. *For more details, please download the PDF or contact us.

• Other Software