←
Property inspections made easy
Industry
Software
Use Case
Home Services
Company size
51-200 employees
Headquarters
San Diego, California
Data labelling
Discover our solutions
About
Yembo’s mission is to digitally transform the home services industry.
Yembo’s founders recognized that visiting customers’ homes to make estimates and inspections is an expensive, inefficient, and laborious process for moving companies. And so, they came up with a solution: AI-enabled virtual surveys.
Yembo’s first product has helped moving companies reduce the need for home visits. Yembo’s clients can take short videos with their phones—and the AI automatically detects objects, creates a visual inventory, and much more. This approach reduces human error while protecting the bottom line by increasing surveying speed and decreasing staff needs.
Thanks to early successes with this solution, Yembo decided to fine-tune the product and build a 3D reconstruction tool. It allowed them to extend their offering to insurance companies, helping them speed up underwriting inspection and claims adjustment. The AI-powered tool provides instant insights into property contents, structure, hazards, and measurements—enabling insurance professionals to move faster, more accurately, and with fewer operational costs.
The challenge
For moving and home insurance companies, visiting customer houses for estimations and measurements is a huge strain on resources. To solve this problem, Yembo has developed AI-enabled visual surveys for property inspections based on user-captured videos. The solution has drastically reduced the need for site visits.
To build their AI product, the Yembo team must annotate a large dataset of lengthy videos with masks for segmentation models. Their requirements exceeded the capabilities of the in-house tool they had previously used for bounding box annotation. Thus began the search for a training data platform with top-class video annotation and segmentation features.
The solution
Yembo chose V7 for its market-leading video annotation toolkit and intuitive UI. Leveraging V7’s video labeling features, mask annotations, and customizable workflows, they trained segmentation models that allowed them to improve and extend their offering. With their updated product, Yembo covers a growing list of use cases across the home services industry—including content inventory, measurements, 3D models, floor plans, structure identification, and more—all of which previously required site visits.
Interpolation, SAM-powered auto-annotation, and easy UI let labelers fuel both the speed and quality of the Yembo team’s work. They also benefit from customizable workflows that allow them to build repeatable training data preparation processes. All of that contributes to faster product development time and higher scalability.
The results
Yembo is pioneering the home services tech industry and is determined to continue moving the needle.
In conversation, the Yembo team revealed that V7 made it possible for them to add their segmentation-based 3D reconstruction tool to their toolkit. Without V7, they wouldn’t have been able to build and maintain a suitable video annotation and segmentation solution in-house. By choosing V7, Yembo instantly accessed state-of-the-art technology and extended its offering fast enough to drive innovation across its industry.
Yembo continues to improve its existing tools while developing new solutions that give customers an even better understanding of spaces without needing to visit them. V7 is proud to support them in building powerful AI that makes property inspections much easier, helping home services companies and their customers save time, money, and effort.
Training data needs
Yembo’s AI tool helps home services companies understand their customers’ spaces without the need for in-person visits.
Yembo developed an object and scene recognition AI tool that analyzes user-captured videos. However, to achieve even greater precision, the team has turned to creating a 3D reconstruction tool based on a segmentation model.
The Yembo team must annotate videos with masks and, occasionally, bounding boxes to prepare datasets for their object detection and segmentation models. Their training data preparation pipeline is organized with the help of V7 workflows.
After seamlessly importing data into V7 with the help of integrations, the Yembo team applies stages (such as logic) in their workflows to filter out the most high-quality and relevant images.
Once the data is chosen, Yembo’s external annotators start labeling videos. Initially, they performed all annotations manually and imported external models with V7’s BYOB (Bring Your Own Model) feature to perform model-assisted labeling. Now, with the integration of SAM (Segment Anything Model) into V7’s auto-annotation tool, the whole process has become even more convenient.
After all the data is annotated, it’s moved to the review stage. Finally, the team easily exports the data to train their model in-house.
Yembo has built a repeatable, scalable, and transparent training data preparation process that allows for higher project observability, more accurate annotations, and easier cooperation with external contractors.
Why V7?
At the beginning of their journey, Yembo used an in-house solution to annotate images with bounding boxes. However, they soon realized that:
They needed a video annotation solution—at first, the team parsed videos for frames and annotated just a few of these frames, which took an enormous amount of time while compromising accuracy.
They needed to train segmentation models to enable 3D reconstructions—but their in-house labeling tool couldn’t support the mask annotations necessary to achieve the desired accuracy, and the team didn’t want to waste resources and talent to develop and maintain new solutions.
Hence, the search for the perfect training data platform began—leading the Yembo team to V7.
V7 met Yembo’s needs for segmentation and video labeling. After taking the platform for a test ride, the in-house team and their annotators were amazed by V7’s intuitive UI and smooth annotation experience. The video labeling toolkit let them annotate videos in their native formats without the need for parsing, and the interpolation feature plus model-assisted labeling made it easy to achieve the desired accuracy in an even shorter time.
Labelers found V7 easy to use and navigate—which allowed them to increase the quality and speed of their work and take on the roles of reviewers. The in-house team fine-tuned their project management and review processes thanks to flexible workflows. Additionally, the S3 integration allowed for seamless integration of V7 into Yembo’s machine-learning pipeline—the team can easily import and export their data.