AI Algorithm R&D Support

We provide support for research and development, such as treatise survey and technical survey that form the basis for research, and algorithm-based research and development and optimization.

Understanding/explanation of individual papers from functional implementation and re-experimentation of paper content, and investigation of global technology trends.
1.We pursue the content of the paper in depth, including functional implementation of the paper content, re-experimentation, and performance evaluation using the customer's data set.
2.We can provide services upon request, ranging from research on global technology trends to understanding/explaining the contents of individual papers.
This detailed research and comparison allows us to determine the advantages of each method and to recommend the method that best meets your objectives.

1.Technical surveys of model compression framework.
2.ML/DL treatise survey, in-depth technical survey, and trend survey.
Image processing:object detection, instance segmentation, posture estimation, anomaly detection, SLAM
Other:NLP, NAS
3.Survey model compression technology and hardware technology information, and summarize devices capable of model compression.
4.Development of a module to evaluate the techniques of each paper in a unified manner.

Operating System	Windows / Linux
Development Languages	Python / C#
Frameworks	PyTorch / NNCF / OpenVINO / TensorFlowLite / TensorFlow

In addition to supporting the research and development of image recognition algorithms, we will also investigate and implement model compression techniques (the purpose being to mount them on devices while maintaining accuracy).

Research and development of new algorithms and optimization of existing algorithms based on the latest treatise information, etc.
By developing and evaluating prototypes with an eye toward commercialization, we can provide support for commercialization in the shortest possible time, including optimization with borrowed data.

1.Research support for ML/DL model compression technology to obtain detailed quantization analysis technology.
2.ML/DL model performance evaluation to improve performance of proprietary models.
3.Implementation for pipeline execution using multiple models.
4.Implementation and evaluation of algorithms for visual inspection, etc.
5.Implementation of a queueing simulator for factories.

Operating System	Windows / Linux
Development Languages	Python / C#
Framework(DL)	PyTorch / NNCF / OpenVINO / TensorFlowLite / TensorFlow / ONNX
Library(ML)	OpenCV/Python library (mainly available on GitHub)

Annotation is required for learning with unique images instead of using public datasets. We create annotations for learning datasets needed for deep learning, such as person detection, object detection, pose estimation, and region selection.

Research and development of new algorithms and optimization of existing algorithms based on the latest treatise information, etc.
1.Perform annotation tasks required by ML/DL, such as assigning rectangles to images, coloring specific areas, etc.
2.To ensure quality, we work with the customer's specifications for increased proficiency and perfection.
3.The work is conducted at the Minato Mirai head office and is done in a security-conscious manner.
4.A system of cross-checking by multiple people eliminates individual differences among annotators and ensures uniform quality.

1.Rectangle for human body detection
2.Boundaries for segmentation
3.Labeling and other annotations