How it Works
LOST was especially designed to model semi-automatic annotation pipelines to speed up the annotation process. Such a semi-automatic can be achieved by using AI generated annotation proposals that are presented to an annotator inside the annotation tool.
LOST is flexible since it allows to run user defined annotation pipelines where different annotation interfaces/ tools and algorithms can be combined in one process.
LOST is shipped with a preset of annotation pipelines for out-of-the-box AI support. If you want to design your own fancy AI pipeline, LOST will provide all the building blocks you need.
It is web-based since the whole annotation process is visualized in your browser. LOST allows to organize label trees, to monitor the state of an annotation process and to do annotations inside the browser.
All annotation pipelines will be visualized in your browser to monitor the state of your pipelines or to start a new one. A pipeline consists of different elements that define your annotation process. Possible elements are datasources, scripts, annotation tasks, visual outputs and data exports.
Our Single Image Annotation Tool (SIA) was designed to annotate single images with Points, Lines, Polygons and Boxes. To each of the above mentioned annotations a Class Label can also be assigned. SIA can be configured for each annotation pipeline, in order to enable specific annotation types e.g. boxes or to prevent different user actions.
The Multi Image Annotation Tool (MIA) was designed to annotate clusters of similar objects or images. The idea here is to speed up the annotation process by assigning a class label to a whole cluster of images. The annotators task is remove images that do not belong to the cluster clicking on these images. When all wrong images are removed, the remaining images get the same label assigned by the annotator.
Python Script API
Define your own scripts that are part of an annotation pipeline. Write any code you like and communicate with lost over the provided python3 API: Train your own machine learning models; Export your data in any format you like; Generate annotation proposals…
Label Tree Management
LOST provides a tree based label management. When configuring an annotation task, specific subtrees can be used as possible labels. Label trees can be imported and exported in csv format.
Workers are docker containers that have the ability to execute LOST scripts. You can start as many workers as you like to distribute the workload among different machines in your network. LOST will visualize the status of all connected workers in the web GUI.