Features and capabilities

3D tool

A 3D point cloud is considered to be a more complex type of annotation, because it is already about 3D. From some types of cameras, information comes in three-dimensional scanning of space and not in the form of videos and pictures. Basically, we are now talking about Lidar - this is a radar, a device that receives a three-dimensional description of objects. This is a slightly more complex annotation level.

High performance video annotation

With the Keylabs platform's technical and software capabilities, video annotation is highly accurate (precision of up to 99,9% depending on project needs) and fast. Thanks to the geolocation adaptation of servers, even big-sized videos are loaded and processed quickly.

Magic wand

Speeds up the annotation process by automatically detecting closed shapes of the same color or color gradient in a highly precise manner.

Object interpolation

Object interpolation in the data annotation process is a handy technique used to accelerate the annotation of objects across a sequence of frames in video annotation.

Here's how it works: annotators label the shape of an object in the first and the last keyframe of desired sequence and the object interpolation algorithm automatically generates the labels for the object in the intermediate frames.

This technique saves time and also ensures consistent labeling across frames. Essentially, object interpolation bridges the gap between initial and final frames, smoothing the overall data annotation process.

A-Z order

Objects can be placed on different leveled layers, which allows operators to correctly detect and work with those objects and their boundaries.

Multilayer annotation

A 3D point cloud is considered to be a more complex type of annotation, because it is already about 3D. From some types of cameras, information comes in three-dimensional scanning of space and not in the form of videos and pictures. Basically, we are now talking about Lidar - this is a radar, a device that receives a three-dimensional description of objects. This is a slightly more complex annotation level.

Object linking

Object linking in the data annotation process is a valuable function that connects different instances of the same object across multiple frames or images. Essentially, it establishes a relationship between the same entity appearing in different parts of a data sequence.

For example, in video annotation, an object appearing in different frames is linked throughout the video, ensuring the continuity and consistency of the annotation. This function is particularly useful in tracking movement or changes in an object's state over time. With object linking, the accuracy and coherence of the annotated data significantly improve, making it highly beneficial for training more precise and reliable machine learning models.

Hierarchical atributes

The attribute is a type of tag that can be applied to a class or property to provide metadata about it.
Using attribute hierarchies, it is possible to define structures of metadata for each item in dataset.
It is achieved by using dependent attributes, which allows logical forming of metadata information for frame or object individually.

Workflow and task distribution

Workflow includes custom stages of one of 4 project stage types: annotation, verification, miscellanious and final.

On the other hand, task distribution is about assigning specific tasks to different team members based on their skills and the requirements of the project. Good workflow and task distribution ensure that the data annotation process is smooth, efficient and completed within the required timeframe. It also helps in maintaining the quality of annotated data as each task is performed by the most suitable team member.

Data management

Data management in the context of the data annotation process is about strategically handling and organizing the data throughout its lifecycle. From the initial phase of data collection, where raw data is sourced, to preprocessing, where it's cleaned and formatted for annotation, each step requires careful management. Further along, it involves controlling the annotation process itself to ensure that tasks are completed accurately and within the set timeframes.

Additionally, it includes maintaining a record of the annotated data and its corresponding metadata for future reference and use. Effective data management helps to uphold data integrity and ensure that the final annotated data is accurate, consistent and ready for use in AI and machine learning projects.