Working With the Client to Automate Last Mile Delivery

Semantic Segmentation

Mindy Support engaged in the meticulous process of annotating images using the technique of semantic segmentation. Within the spectrum of annotations, various elements required attention, encompassing pedestrians strolling along walkways, the intricate network of roads, the pathways in the form of sidewalks, and the crucial demarcations of crosswalks.

The client was developing a last-mile delivery robot that would operate on pedestrian sidewalks and crosswalks using autonomous navigation in compliance with traffic regulations. Its navigation system is equipped with multiple RGB cameras that identify pedestrian pathways and crosswalks to ensure safe navigation. The labeled dataset collected will primarily train the AI model to distinguish between sidewalks and crosswalks from other surfaces like roads, grass, sand, mud, etc. This distinction is crucial in developing an accurate navigation path for the robot.

However, simply having this dataset was not enough. For the robot to accurately learn to recognize all of its surroundings in the physical world, the images needed to be annotated with semantic segmentation, which is one of the most detailed types of data annotation. More specifically, all the pedestrians, roads, sidewalks, crosswalks, and other objects of interest in the images needed to be annotated with semantic segmentation. 

The client was looking for a reliable data annotation provider that could annotate the training dataset with a quality score of 98%+.

The client issued a tender for prospective data annotation providers. We created a pilot project to display our expertise and how quickly we can grasp the inner workings of their engine and platform. The client was very satisfied with the results and awarded the contract to Mindy Support. 

The entire dataset consisted of 3,000 images, and the client had a strict deadline of 2 months for all the annotation work to be completed. We assembled a team of 10 data annotators to get the job done. We started out by training them on how to use the client’s tool and how the semantic segmentation work needed to be completed. Since our data annotators had extensive experience with this type of annotation, it did not take them long to grasp all the concepts and requirements. 

Our QA team was making sure that all the requirements were being followed and that we were on track to reach the quality score required by the client. We were able to exceed the client’s expectations by getting all the data annotation work done ahead of schedule without sacrificing the quality of the work. The client was very pleased with the results and planned to further expand the project. 

• 3,000+ images annotated 
• Assembled a team of 10 FTE
• Reached a 98%+ quality score