Utilizing the company’s AI advantage, 4 members from FPT Software had come up with an AI Labeling product, which shall help reduce labeling time for human, vehicles… on images of various sizes, and currently had successfully developed the product’s first phrase. The application holds strong promises for wide application in various aspects of life.
AI Labeling is developed based on integration of various image processing and object recognition technologies like Deep Learning and Machine Learning. Moreover, AI Labeling is being integrated with Reinforcement Learning (RL) technology – a combination of synthetic neural networks and reinforcement learning structures, which allows users to improve the accuracy of the AI Labeling app as time passes.
The team had completed the first phrase in app development at the end of last May. With this, AI Labeling app can now identify human, motorbikes, buses, cars, and trucks on images of various different sizes, as long as the images retain critical features of the aforementioned objects. Furthermore, AI Labeling can also replace core AI models so that the app can be used most suitably for specific data types.
In the second phrase, the app will expand labeling support, not only on rectangular frames, but also on polygons, which will be a great help for AI projects on segmentation data labeling. This also allows users to review previous labeling results, add or delete unnecessary objects, and change shortcut keys per preferences. Along with that, with the automatic labeling function, the app also inherits features of other popular devices, thus increase convenience and familiarity for users.
In comparing AI Labeling with other similar applications, Mr. Dinh Xuan Tue – the project team’s representative, had said: “Current labeling tools are often offline applications, which take time for installation, and may occur errors based in the users’ device environments. However, AI Labeling Tool is developed based on web-based techniques, where users only need a computer with Internet connect to access the server and start the app, without having to care about library installation or working environment.”
At the moment, to use AI Labeling, users will need to connect to FPT Software’s internal network. This is a minor problem which reduces the application scale. In the next phrase, the team shall develop to bring AI Labeling on cloud so that it can be used from anywhere. They also plan to expand the types of labeling frame, from rectangular to other polygon forms, in order to improve the application.
In the near future, AI Labeling will aim at getting more diversification functions for label data, particularly from just integrating images, to other forms of data like sound, video, and text.
More details regarding AI Labeling can be viewed in the video below:
Source: FPT SoftwareRelated posts: