Intel CVAT, a handy open source data annotation toolkit

As we know, the training of deep neural networks requires huge amounts of data. In addition, depending on the AI ​​architecture you are holding, that data will not be used repeatedly if it is not labeled, this will take time - especially when you are deploying projects with scale up to hundreds of thousands of objects.

To reduce the burden on data annotation experts as well as data scientists, Intel has released a new program with open source, which is a computer visual annotation tool (Computer Vision Annotation Tool). - CVAT), built to help simplify and expedite the annotation process of video and image samples used to train computer visual algorithms. Intel has announced in their blog post as follows:

'To accelerate the process of annotating data in the Computer Vision field, we have developed a program called CVAT. In general, there are many ways for you to annotate data, but using special tools like CVAT can make this process simpler and faster. "

Intel CVAT, a handy open source data annotation toolkit Picture 1

1. Wireshark 3 is officially released, comes with the new Npcap Windows driver

As Santa Clara Data Management Company explained earlier, CVAT will deploy via Docker and be accessed via browser-based interface (or embedded with options on platforms like Onepanel), copper Time to own a task-based management system optimized for specific tasks (users create public tasks to divide work together). In addition, CVAT also supports supervised machine learning tasks related to object detection, image classification, image segmentation and annotation with 1 of 4 types: box, polygon, multi-stroke and multi points.

In CVAT, annotators contain a lot of tools to copy and spread objects, apply filters, adjust visual settings, make annotations automatically, or detect objects through APIs. Google's TensorFlow framework . In addition, CVAT will also perform a number of automated (but not all) tests, and has unique compatibility with data analysis tools like Intel, OpenVino, Nvidia, Cuda and ELK (Elaticsearch, Logstash and Kibana).

Intel CVAT, a handy open source data annotation toolkit Picture 2

1. Researchers use computer vision to better understand optical illusion

It can be said that CVAT is a combination of many jobs in a process. Intel notes that this toolkit has only been extensively tested with Chrome, and is not recommended for use with Chrome Sandbox (the "sandbox" feature in Chrome helps limit the environment for processes, including RAM usage and negatively affect the performance of this toolkit). However, Intel also pledged to gradually improve this weakness of CVAT in the future.

'CVAT is built for professional algorithm and annotation development teams, and also receives great support from these people, and we have tried to provide the most useful features. User feedback will help Intel better define the direction for future CVAT development. We hope to soon improve the user experience, feature set, stability, automation features and compatibility with other CVAT services. At the same time, Intel also encourages interested people around the world to actively participate in developing this toolkit. '

Intel CVAT, a handy open source data annotation toolkit Picture 3

1. New technology beyond helps sign language barriers

Intel decided to release CVAT only weeks after Uber officially opened its automated visualization system (a web-based platform for media data), and when startups also operate in the field. Data annotations such as Scale and Hive began to increase venture capital and attracted large customers such as Uber and General Motors.