Keymakr - Frequently Asked Questions (FAQ)

A Project management team calculates deadlines and communicates them to the client.

Sure, this is possible and easy to do with our annotation platform Keylabs. Some slight adjustments for your custom format might be needed though. Technical team will easily take care of it.

Ofcourse, this is a part of our annotation pipeline, not only we can provide you with ETA but also elaborate and point out bottle necks which may occur in the process and suggest how to annotate your data in the most efficient way, based on our experience with other clients from the same industry.

We are working with 3 precision levels for both, image annotation and video annotation:

• Precision level 1 is applied when detailed, high accuracy annotation is required. The labeling is affixed with a 1-pixel (or less) spacing at this level aka pixel perfect annotation.
• Precision level 2 is applied when a detailed annotation is needed, but some deviations are acceptable. It is usually used when the optimal balance between speed and quality is needed. At this level, the labelings are marked with a small gap of 2-3 pixels between the objects and the annotation boundaries.
• Precision level 3 annotation involves even less precise annotations without detailed object labeling.

Please take a look at our precision level documentation slides to learn and see our annotation precision level in more detail: guide.

Sure, as mentioned before, Keylabs annotation platform allows us to adjust the project settings on the go, so the bitmask annotation can be easily converted to polygon.

Yes, Integration with different data pipelines is possible using Keylabs APIs or SDKs.

All of the mentioned methods are supported. But S3 is most preferable.

Absolutely. It is possible to import pre-annotation to your project.

Sure, we fully support this, you can also merge the movies back if required.

Example: A project may specify that some key points of a "skeleton" should be visible, but they may be invisible according to human judgment.

A human judgment factor arises when a certain annotation property/attribute, like occlusion, blur, visibility, or accuracy, can be interpreted differently by different people.
In order to solve such problems, we use the following methods:

1. Prevalidation of client manuals by our project managers
2. Keeping the client informed of potential edge cases through a direct line of communication
3. By mutually agreeing on rules, we can make annotations consistent
4. Before moving forward with the full scale of the project, test materials are always created to ensure expectations alignment

Yes, This is a special feature of our Keylabs platform. Keylabs platform allows you to keep hierarchical (parent-child) dependency between any objects. This can be done during the annotation process.

Yes, This is process is called annotation materials preparation, we can even collect the materials for you if you do not possess them. This process is evaluated and done as part of our annotation services.

Of course. This issue is handled by our IT team. The map is automatically created when materials are uploaded to the project.

In Keylabs' annotation platform any applicable object can be exported as a bitmask, which means the segmentation marking of the object. You can always view the annotation over the image in the annotation editor using your client platform user.

ABSOLUTELY NOT. Your data is YOUR DATA. It is NEVER shared with anyone except you and our in-house annotation team that is working on it.

We have a standard data retention policy for 2 month after project completion, of course this can be changed to any direction. In general our customers request longer retention periods due to the life cycle of ML projects.

Where is my annotation project’s data stored?

1. Default option : The data is stored on our dedicated servers which are located in our secured facilities
2. Second option: Data can be stored on aws S3 or EBS storage
3. Third option (customizable): Data can be stored on clients servers or any other hosting provider with or without VPN

Bounding boxes: axis aligned or rotated - This is the fastest annotation technique, and also the most common. Using our Keylabs annotation platform we drag bounding boxes around objects.

Cuboid annotation - adds more dimensions to training image datasets. We create cube shapes by using a 2D box to locate one side of an object, then an additional box is used to identify the opposite side.

Polygon annotation - this annotation type is essential if developers are looking to capture irregular or complex shapes.

Skeletal annotation - this technique allows AI models to identify and interpret the bodies and movement of humans.

Key point annotation - This technique is used to label important, single points in images.

Lane annotation - linear and parallel shapes and structures are traced using this technique.

We got something to read for you https://keymakr.com/image-annotation.pdf

For project price estimate, we need to run a pilot on your data to measure the time and effort needed for the annotation. Every project is different and requires different types of annotation, therefore we must run the pilot for correct estimation.

Sure, please visit https://keymakr.com/partners.php and fill out a registration form.

Absolutely Yes! Our formats are optimized to work both on video and image annotations, creating a lightweight, easy to read format that covers all the possible annotation requirements. Our format has been tested and proven to be easier for handling, by many of our clients.

Merge is a simple function when two sequential video movies are being glued together. This function allows you to cut a single long movie into several pieces and then stick them back together while preserving the continuation of object tracking ID. In simple words if the same person appears on the video from both sides of the merge “stitch” it will be merged to be one object in the merged video preserving its ID.

Yes, it’s possible as an optional request.

Yes, we do have a CLI and a SDK.

We easily support thousands of objects in one single image.

Sure, we are using Keylabs annotation platform, which is highly customizable and allows us to do this by adjusting the project settings.