Why We Don’t Support Cloud Recognition
Cloud recognition is appealing as it allows developers to store a large amount of markers without bloating their apps size. The trade off is that your image detection will suffer from network latency. Even with a low network latency the performance of your marker detection is going to be significantly worse than if your marker was stored locally. This results in a frustrating user experience as the time spent looking at a marker is greatly increased. In order to achieve an acceptable level of network latency developers would need to ensure they had servers close to wherever their application was being distributed, which is expensive.
An alternative suggestion to cloud recognition is to temporarily download and store the markers for a given situation such as a users location. Updating markers using this method will remove any network latency issues from marker detection, whilst still providing developers with the flexibility to update their applications content without having to alter the underlying source code. The Kudan AR Toolkit provides developers with the ability to generate KARMarker files, which can store a large amount of markers in one lightweight marker set. This is useful when downloading and updating the markers you application uses as images do not need to be downloaded one at a time and names can be set automatically.
We have a simple example that demonstrates how to load markers and augmentations from a downloaded path here: https://github.com/kudan-eu/CMS-Demo-iOS