NTT COMWARE to Launch a New Version of its Deeptector® Image Recognition AI
- Accelerating companies' DX through higher detection accuracy and learning efficiency -
NTT COMWARE CORPORATION (Head office: Minato-ku, Tokyo; President: Satoshi Kurishima; hereinafter, "NTT COMWARE") announced today that on October 13 it will start offering a new version of its Deeptector® image recognition AI featuring a renewed architecture and its new area detection function.As companies promote digital transformation and workstyle reforms, demand for operational reforms leveraging AI has grown. However, the use of AI requires technical knowledge. In addition, the creation of an AI that can be put into practical application requires time-consuming learning tasks and accuracy improvements. This has hampered the application of AI in actual operations, which is an issue. NTT COMWARE has been verifying segmentation technologies for image recognition AI with companies in the telecom and construction industries, among others. With these initiatives, the company has confirmed these technologies are accurate enough to be applied in operations and are an effective means of improving the efficiency of learning tasks. At the same time, the company has enhanced the Deeptector GUI client tool's functions that visualize detection accuracy, which support any user's easy AI creation. In connection with this, the company will start providing area detection as a new service.
Further, NTT COMWARE has redesigned the platform architecture of Deeptector using container technologies. This enables users to easily add and change deep learning (DL) engines which are optimized for particular operational issues, and DL engines they have developed on their own, in addition to the standard Deeptector DL engine, while the use of containers shortens the development cycle. Also, the standardized application programming interface (API) has made changing the application unnecessary when the engine has been changed, allowing corporate users to provide new services using Deeptector more promptly and efficiently than ever before. The renewed architecture enables users to incorporate the latest AI technologies, which continue to evolve every day at an accelerating rate. It is expected that this new Deeptector will enable companies to quickly apply the latest technologies to their operations and accelerate digital transformation.
NTT COMWARE promotes the use of AI through the continued enhancement of Deeptector and other AI services, thereby helping companies to achieve digital transformation.
Deeptector is an image recognition AI using deep learning. The AI solves operational issues by acting as an alternative to human eyes in inspections, tests, and similar operations. Deeptector provides a range of functions necessary for the use of image recognition AI — the creation of teacher data, learning, and detection — via its GUI client tool. It enables users to create an AI using four detection functions: object detection, classification, level detection, and defect detection based on images of non-defective items. The new version features the new area detection function that uses segmentation technologies, a renewed platform architecture, and enhanced GUI client tool functions.
2.Features of the new version
(1) Area detection function using segmentation technologies
Area detection is a function that detects an object, even identifying its shape. The shape is learned by surrounding the object with a polygon that follow the object's shape. Because even the shape of the object is identified, this function minimizes noise caused by the background or overlapping objects. Further, users can expect accuracy improvements even if they lack technical knowledge because of the default parameters that are set.
(2) Redesigned platform architecture
The redesigned platform architecture enables the following.
- Use of containers to significantly shorten the development cycle, from testing to commercial use
- The user can easily add new DL engines, in addition to the standard Deeptector DL engine.
- The standardized API has made it unnecessary to change the application after the addition of a DL engine
- User-defined image pre-processing and post-processing tasks can be added (hook function).
- Learned models created using different DL engines can be used in combination.
(3) GUI client tool supporting accuracy improvements
The Deeptector GUI client tool comes with enhanced reporting functions for the visual confirmation of detection accuracy, including graphs and an aggregating function. It enables the creation of teacher data, learning, and detection, which are necessary for the creation of an AI, all from one service, thereby supporting the time-consuming tasks necessary for improving the accuracy of AI.
Figure 1. Visualization of the distribution of the number of detections
Figure 2. Narrowing of falsely detected images
To use the area detection function, you need to sign an optional contract (for value).
4.Date when the new functions will be available
Tuesday, October 13, 2020
*"Deeptector®" is a registered trademark of NTT COMWARE CORPORATION.
*Other names of companies, products, and services are trademarks or registered trademarks of their respective owners.