🎥 Powerful video labeling tool for Deep Learning: training data for Computer Vision with Supervisely

There are a lot of applications and industries where it is needed to apply Neural Networks to videos: agriculture, self-driving cars, robotics, manufacturing, consumer electronics and so on.

But 80% of existing tools are closed and are not available for the Deep Learning community, the other 20% — are not designed for annotation at scale.

We at Supervisely are proud to announce the web-based Video Labeling Tool. Community Edition (CE) is free for research purposes and we hope the entire Machine Learning community will benefit from it. Contact us to learn more about Enterprise Edition (EE).

Video Labeling Tool — user-friendly and intuitive

• inspired by professional video editing software
• created by data scientists for data scientists in collaboration with labeling workforce
• fast and web-based
• great user interfaces with dark theme
• customizable view settings (brightness, contrast, …) and hotkeys for everything (navigation, labeling and tagging)

Supervisely Video Labeling Tool: interface overview

Scene settings and Hotkeys

AI object tracking out of the box

• state of the art class-agnostic Neural Network for object tracking is integrated into video labeling tool. Can be deployed both on CPU and GPU
• custom tracking NN can be integrated using REST API
• OpenCV tracking, Linear and Cubic interpolators are also there

AI-powered video segmentation with SmartTool — fast and accurate

• we’ve optimized user experience to segment any objects on videos with a few clicks
• the class-agnostic model can be used to segment any object. A training procedure is available in Enterprise Edition (EE).

AI-powered video segmentation

Video Timeline for easy navigation over thousands of frames

5-minute 24 FPS video contains 7200 frames. Imagine, how many frames a 2-hours video has. Video Timeline is designed for

• quick navigation: you can find frame ranges with or without labels
• operating with ranges: play video segments, select a range to delete objects or assign tags

Video timeline example

Multi-camera video labeling

Now you can label a single object that is presented on different videos. Video surveillance or self-driving cars are good examples.

Label objects on different videos

New rendering engine

• super-fast rendering
• each labeling tool has its own custom behavior: the creation of polygons, rectangles, bitmaps, cuboids, and landmarks became more intuitive and user-friendly.

New polygonal and rectangular labeling tools

Main labeling concepts: frame, object, figure, tag

• a single object consists of figures labeled on multiple frames
• assign Key-Value tags with optional frame range to objects and videos

Object and its figures on different frames

Natively supports 4K and any video format

• Enterprise Edition (EE) supports avi, mp4, 3gp, flv, webm, wmv, mov, mkv and more …
• realtime video streaming and transcoding on the fly, thus no need to convert and store data in different formats
• has built-in tools to connect existing video storages: add your private local storage or the cloud one (Google Cloud, Amazon AWS, Microsoft Azure)
• working smoothly with high-resolution videos for accurate labeling

Integrated to the entire Supervisely ecosystem: labeling jobs, neural networks, API, SDK and more …

Having the video annotation tool is only half the battle. We integrated it into the main platform, so you can benefit from other cool features:

• user management and access permissions
• Labeling Jobs to work with labeling workforce
• integration with HTTP-based RESTful APIs and Python SDK: add your custom Neural Networks and other plugins
• use our built-in dual-pane file manager DataCommander to manipulate data

All the small things

• cursor hints and special keyboard shortcuts

Pro-Tip: how to quickly create a new object or add new figures to the existing one

• class “Any Shape” —label objects of the same class with different instruments like polygon, rectangle or brush if you want.
• tune image resolution in labeling interface — useful if you or your labeling workforce have a slow internet connection
• additional tags statistics in labeling jobs
• copy project classes and tags from one project to another in a few clicks
• download labeling activity as a CSV file
• autocomplete fields during labeling jobs creation
• new user role — “Reviewer”
• NNs plugins support both CUDA 9 and CUDA 10
• additional tutorials and integration guidelines with examples have been added

Try it now!

It’s already there — just import videos using our built-in Videos plugin. If you don’t have an account in Supervisely yet — sign up for free now.

Feedback is appreciated — it will help us to build a better product

If you found this post interesting, then let’s help others too. More people will see it if you give it some 👏.