Challenge
Scythe Robotics’s autonomous commercial mowers operate on large lawns. They are equipped with stereo camera pairs and other sensors.
The objective of the perception model is to recognize fully drivable area and distinguish between an area covered with leaves or dirt, lawn borders, and unknown objects.
Scythe takes a deep learning approach, which requires a considerable amount of accurately labeled 2D and 3D data across a variety of more than 50 classes.
Solution
Scythe segmented drivable areas and obstacles on the platform for the past three years to create accurate labels.
The custom insights tab enables the engineering team to follow up on the most critical labeling metrics,and billing of the outsourced labeling.
In the past years, the Scythe team and Segments.ai collaborated closely to develop and improve numerous features to speed up the labeling, such as SAM.
Benefits
Manage your labeling team
The management tools on the platform allows you to onboard your own outsourced labeling team, while keeping a close feedback loop on quality.
Faster labeling with ML-assisted tools
Workflow metrics at a glance
Integration with the existing pipeline
The integration with the in-house tech stack allows a streamlined flow of data.
Scythe Robotics at a glance
The relationship with Segments.ai has been great, with great support. The tools used for the labeling process are simple to use, with improvements making the experience better every day.
Jose rendon leyva
Data Operations
Scythe Robotics
Segments.ai’s Superpixel tool for image segmentation boosts our labeling efficiency significantly. The workflows, Python SDK, and webhooks have been game changers for us. The platform is now seamlessly integrated with our active learning pipeline.
Andres Milioto
Senior Computer Vision Engineer
Scythe Robotics Alumni
Author of SemanticKITTI and RangeNet++
