Through the years, the Ximilar team has worked with many image annotation tools. Some of them were desktop-based and some of them web-apps. Most of the desktop applications are offline and intended for single person use, not for teams. So it is a challenge when your team is growing and your project requires higher complexity. Scalability becomes a real issue here.
We also understand well, that building a solid annotation tool for computer vision is not an easy task. And yes, it requires a lot of fails and exploring many wrong directions, in order to find a better solution. So let’s see if we made it.
Fighting the Challenges
The Computer Vision Platform (Ximilar App) that we are building, can be used in multiple areas. We have teams working in domains such as Fashion, HealthCare or Indsutry40. That is why we need a tool pretty complex to cover all of the requirements like:
- Collaborative work for teams
- Solid user experience → Everything should be made as simple as possible, but no simpler
- AI helpers and tools should lead to the high-quality datasets
- Building image annotated datasets fast
- Fast prototyping of new features
- and more
That is why we have created annotation tool annotate.ximilar.com. And the best part is, that you can now use it too!
The annotate.ximilar.com is connected to the same backend & database as our Ximilar App. So the images and labels uploaded through the API or app.ximilar.com become the same entities. However, Annotate extends the simple view of Ximilar App to many more advanced use-cases.
The preferred view in Annotate is usually a single image view. Because you are allowed to draw on the image and therefore be more focused on it.
Why even verify your data?
Annotate helps you build high-quality datasets by verifications. Building good computer vision models requires a lot of data, high-quality annotations and a team of people who understand the process of building such a dataset. Every image can be verified by workspace users. You can train the models only on verified images and increase the precision. If you are looking for high-quality models then you need to understand your data and have a perfectly annotated dataset. In the words of Andrej Karpathy (Director of AI at Tesla):
Repetitive Tasks Automation
Annotate can speed up the work of your team, to build complex vision solutions.
- Creating annotation jobs and split the work among your team members
- After your tasks are initially trained, your team is allowed to get a fast prediction on not processed images
- Create bounding boxes for object detection with one click
In some cases, recognising the most probable category or tags is not enough. Sometimes you need to know the location of the specific object. Through Annotate, you are able to create Detection Task. This task is trained on the bounding boxes drawn on the image. Such a service will be explained in another post on the Ximilar Blog in more detail. We are working also on image segmentation service, so stay tuned!
Benefit → Speed of your Team
We will show you all the technical features of Annotate in the next blog post. Annotate is still a beta at this moment, but we hope that you will love the tool. If you are working on a complex project with a team, you can improve your vision models unbelievably fast — just let us know and we will active the Annotate tool for you.