Log in or register to post comments

Slow Image Target detection for large dataset (50+ targets)

September 28, 2020 - 5:52am #1

Vuforia SDK Version: 9.1.7 -

Description with steps to reproduce: Hello,

I am using Vuforia in Unity with a local database containing 140 Image Targets, each of which has a 5-star image target rating. Since updating Vuforia from 8.3.8 to 9.1.7, tracker detection has become very slow on Android devices. Sometimes it takes 10 to 20 seconds before the target is detected.

I've tested with printed targets in different light conditions and off my pc monitor, with the same results.
This is also tested on multiple different Android devices, including high-end phones.
This behaviour does not occur in the Unity Editor using a webcam, or on iOS, both those platforms have fast detection.
This behaviour did not occur when using Vuforia 8.3.8.
This behaviour occurs for both Mono and IL2CPP Android builds.
When reducing the number of image targets in the dataset, detection speed becomes better, but only reaches acceptable speed when going as low as 20 targets. From the documentation I understand that for datasets over 1000 targets, switching to cloud recognition is recommended, so 140 targets should not be a problem right?

I have tried upgrading to Unity 2019.4 and Vuforia 9.2.7, this yields no difference in detection speed.

Do you have any idea why target detection speed on Android is so slow? -

Development OS (Mac OS X, Windows, Linux): Windows -

Mobile OS and Version: Android 9, iOS 13 -

Mobile Device Manufacturer and Model name: Oneplus 3, iPhone 7 -

Do the Vuforia Sample Applications show the same behavior?: Not applicable since the problem lies with the dataset size

Slow Image Target detection for large dataset (50+ targets)

September 29, 2020 - 2:47am #2


Would it be possible to share your 150+ image targets db with us for testing?  please feel free to pm with the download link

Thank you.

Vuforia Engine Support


Log in or register to post comments