Better machine learning models, fast
Machine learning models are only as good as the data they are trained on – garbage in, garbage out. When it comes to getting your training data in the best shape possible, good annotations are key. When it comes to object detection in images, many architectures focus on bounding-box annotation. These low fidelity annotations include extra pixels of non-target information that often result in unintentional biases in your models.
Instance segmentation has long been regarded as the gold-standard solution to these problems, however creating segmentation masks has historically been too slow, expensive or difficult to be a viable option for many.
Zegami Image Annotator is designed to solve this problem: a visual, intuitive and fast way to curate training data sets, create segmentation masks and then integrate into your existing machine learning workflow.
Full integration with the Zegami Machine Learning Suite
Zegami Image Annotator is fully integrated with our Machine Learning Suite. After launching Image Annotator, simply log-in to your Zegami account and start adding annotations to the images in your online collections. All annotations are stored on the cloud alongside the images for effortless collaboration.
Selectively fetch images. Not all images in a collection will necessarily need to be annotated, so with Zegami Image Annotator you can pull in images with a specific tag. Use Zegami’s powerful exploration and filtering features such as ‘image similarity view’ to quickly mark those that need attention.
Review annotations in Zegami. Once annotations have been authored and saved using Zegami Image Annotator, browse the results in Zegami, and tag any that need revising.
Image Annotator benefits
Pixel perfect segmentation
Most annotation tools are tedious to work with. Drawing boxes around thousands of objects makes it difficult to focus and the quality of your training data drops. With Image Annotator creating masks is straightforward: a simple box, a quick adjustment and you’re done. If you prefer drawing polygons or painting masks by hand, then you are free to do so.
Collaborate with Zegami Image Annotator to share the workload, peer review each other’s work and track overall progress of the project. Whether tracking annotations by the person that made them or by class, Zegami makes visualising your assets simple.
Zegami Image Annotator is seamlessly integrated into Zegami Machine Learning Suite. It turbo charges the curation of your training data. Quickly and visually identify data and image quality issues. Subset your data into classes, automatically cluster similar looking images and identify bias. Having this overview of your data saves days of wasted effort when image quality issues are discovered part-way through an annotation project.
How it works
Once the target images have been identified, you can then view and annotate your target images in Zegami Image Annotator. The tool makes it easy to draw bounding boxes, polygons and pixel-perfect segmentations with both manual and algorithm-driven tools. All annotations are automatically synchronised to Zegami so they can be accessed anywhere and by any of your collaborators. The resulting annotations can be exported in a convenient, customisable format that can then be integrated into your machine learning workflow.