> For the complete documentation index, see [llms.txt](https://docs-en.kivicube.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs-en.kivicube.com/ar-asset-creation-guide/image-marker/image-tracking-specifications.md). # Image Tracking Specifications ## Overview ### Scene * **AR Scene: Image AR** * **Collection: Image AR** ### What is Image Tracking? Image tracking, in full, refers to Image Detection and Tracking (i.e., an Image AR scenario). It consists of two components: * Image Detection * Image Tracking
To evaluate the quality of the experience, both factors must be considered: * Whether image detection is fast enough * Whether the AR scene remains stable after detection > ![](/files/mgodJcMPF1l3iTUYUIYg)![](/files/ZHexdCJjYePbWd9ivEw2) > > When using the Collection feature (multi-scene recognition entry), the cloud-based image recognition algorithm will be used. > > Therefore, the recognition images must comply with both the Image Tracking specifications and the Cloud Recognition Image specifications. *** ## Guidelines
CategoryDescriptions and Examples
Valid Formatjpg, jpeg
Valid Color Mode

Set the image to RGB to help detect any color deviation promptly.


When exporting, check “Color Space → Convert to sRGB”, and verify after export to ensure no color deviation is present.

Recommended Image Resolution

The image resolution is recommended to be between 480×480 and 1280×1280, with around 800 being ideal.

Recommended Image Aspect Ratio

Landscape: 1:1 to 16:9 (Aspect Ratio 1 to 1.78)

Portrait: 9:16 to 1:1 (Aspect Ratio 0.56 to 1)

Rich Detail

The following are examples of poor-quality images:


Avoid extensive whitespace

Too much empty space can reduce tracking stability. Minimize blank regions and ensure the main subject is clearly emphasized.

Avoid repeatitive patterns / Symmetrical images

Symmetrical images tend to be unstable:


For images that are not perfectly symmetrical, the final assessment should be based on actual tracking performance.


Repetitive Patterns Can Lead to Detection Difficulties:

Uniform Distribution of Feature Points

Try to avoid large Single-color region around the borders.
Add appropriate details to achieve a more uniform distribution of feature points.

Cloud Recognition vs Image Tracking

Different Feature Point Extraction Principles


As the principles of feature point extraction differ, certain smooth or rounded shapes may receive a low star rating in cloud recognition, yet still perform adequately in image tracking.

Actual Performance may still jitter.




Higher Tolerance for Image Blur:

Even with complex or slightly blurred images, tracking can remain relatively stable

(This does not encourage using blurred designs, only indicates that the system offers higher tolerance.)

Avoid Gradient DesignsAvoid using large gradient areas
*** ## Tips ### How to Improve Image Recognition **Recommended Practice** Before uploading an image for recognition, add a white border around it. If the image will be printed, include a white border on the printed version as well.
### Other Methods to Improve Image Tracking Stability **Place AR objects tightly on the Image marker**
**The AR scene placed on the Image marker should not be too tall.** Example: Scanning a wall poster triggers an AR interaction.
* Users’ perception of model "shaking" is related to its height: the higher the AR object, the more noticeable the visual displacement at the top becomes. * Design AR interactions thoughtfully to reduce the amplified shaking effect caused by overly tall layouts. ### **Include animation in the AR experience** * Rich scene animations can visually dilute the discomfort caused by minor jitter. * In some cases, even when tracking jitter is present, users may hardly notice it.