Start a new topic

Black Oval

We are using Wikitude version 6 and randomly upon recognizing an object and displaying a 3D model we get a black oval on the screen over the top of the 3D model.  We've attached a screenshot of what we are using.  Once you get the black oval it will continue to display for objects recognized until you close Wikitude and go back into Wikitude.  It doesn't require closing the app entirely, just Wikitude.  Any ideas on what needs to be done to solve the problem?


We are actually not seeing the same type of issue on version 7.  However, our recognition is significantly worse on version 7 than it is on version 6.  So much worse that version 7 isn't usable for us.  Any ideas what the difference could be between version 6 and version 7 with the recognition?  Is there any way to get version 7 to recognize as well as version 6?  

Black Oval.JPG
(2.29 KB)

Hello Heather,

Please follow the guidelines provided in the documentation section here.

Thanks
Eva

 

So I set the extendedRangeRecognition to AR.CONST.IMAGE_RECOGNITION_RANGE_EXTENSION.OFF. I just wanted to verify that setting extendedRangeRecognition to that was correct. We are still seeing a lack of recognition with the setting set like this. Also, in this thread it was mentioned setting the carmera to a lower resolutin could improve recognition. Is there sample code for setting the resolution lower? Where would most of those changes need to be made? Would those changes need to be made in the plugin?

We changed the property "extendedrangerecognition"  of SDK 7.1 according to your recommendation, although the performance is still worse than that of SDK 6.1, but it's acceptable. Well done。


Mark Cai

Nice AVI, Eva, thank you.

We'll continue to test the performance between them, because we want to upgrade to SDK 7.1, and we try to disable the property "extendedrangerecognition" when tracking the image, we will report the continuous results. Thank you for your effort.

Mark Cai

Hello Mark,

we have tested your experience in house by using the SDK 6.1 and SDK 7.1, using two similar devices placed directly next to each other (see video below).
Indeed, your target is probably recognized a bit faster with SDK 6.1, but in our oppinion its just minor noticable.

And yes, in order to improve the overall accuracy and stability of the initial camera pose estimation along with featuring multiple targets in SDK 7, we did some internal changes to the algorithm that may affect the perceived recognition duration in some edge cases.
Anyway, we are constantly working on improvements of the recognition and tracking performance of our products aiming at supporting more and more difficult use cases.

Please also note, that in SDK 7 the extended range recognition feature is enabled by default. It enables the targets to be recognized at larger distances, but also requires more calculation effort, and thus, time.
We can advice you to try disabling the extended range recognition using the property “extendedRangeRecognition” of the ImageTracker Class or directly set the Camera to a lower resolution to better match the behaviour of SDK 6. (See https://www.wikitude.com/external/doc/documentation/latest/android/hardwarecontrol.html#camera-resolution)

Hope that helps!

Best,
Eva

 

avi
(5.41 MB)

Hi, Eva,

 

It is embarrassing to take video for us because of software tools and how to describe the question in Chinese (Poor oral English). Our problem is not the picture can not be identified, the picture is recognizable, but the speed of recognition is noticeably slow with new version. SDK 7.1 recognition speed is significantly behind the SDK 6.1.

 

We downloaded SDK Examples (SDK 6.1& SDK 7.1) from github, use the same rmb-100-v5.wtc file, put it under the asset \ samples \ 01_Image $ Recognition_1_Image $ On $ Target \ assets directory, and then modify the recognition object in this example. And then use SDKExamples 6.1 and SDKExamples 7.1 for comparison, it is easy to find the picture recognition performance changes.

 

We have just carried out the experiment, the problem is still very obvious, we are not only one person here to find this problem, three more persons have found the same problem.

 

We sincerely hope that you will be able to solve this obvious problem.

 

Mark Cai

Hello Mark,

We just performed several tests (with Huawei P9) with the image you provided and I could not see any Image Recognition problem. Do you think that maybe you can send me a video where you demonstrate the issues you are experiencing?

Thanks
Eva

 

Hi, Eva,


My platform is Android + javascript (not Native) + SDK 6.1, our mobile brands are OPPO and Huawei which are more popular than iPhone in China.


Our developing environment is Android studio (2.3.3) + Window 7.


Any more info please let me know if you need.


Thank you.


Mark Cai


Hello Mark,

Thank you for sending your targets. Could you also please tell me the platform you are testing with (JavaScript, Native or any extensions) and with which phone (model and OS version)?

Thanks
Eva

 

Hi, Eva,

 

Its easy to find the performance differences between SDK 6.1 & SDK 7.1.

 

If you have the opportunity to try out the picture I gave (rmb-100.jpeg) and the wtc file (rmb-100-v5.wtc from target Mangement which quality with 2 stars, version 5). Using the SDK 6.1 Image Recognition and SDK 7.1 Image Recognition. Look at their performance differences.

 

Our use case is for normal users to take any photo at anywhere to embed a virtual object, and let others find it. Not too hard to find it with SDK 6.1, but now to hard to find use SDK 7.1.

 

Thank you for listening to my nagging.


Mark Cai

rmb-100.jpeg
(135 KB)
wtc
(30.6 KB)
Hello Mark,

Could you please share your use case with us and maybe also a video demonstrating the issue that you are having?

Thanks
Eva

 

Hi, Daniel,

 

Thank you for your quick response.

 

We moved our SDK version from 6.1 to 7.1 directly, our devices are OPPO, Huawei, with the same devices, the upgrade experience is obviously different. Picture recognition function is basic function and very important, compared with SDK 6.1, user AR experience drops very clearly. Maybe the identification error with SDK 7.1 is reduced, but the user experience is also greatly reduced. we moved back to SDK 6.1 now and unable to accept user experience with SDK 7.

 

Thanks,

Mark Cai

Daniel,


I wanted to chime in about our experience.  We are seeing drastically worse results on the 7.0/7.1 release as well.  We finally got someone at Wikitude to tell us that they knew the recognition was lower, but it was done on purpose because there are less false positives.  Essentially, the recognition is so poor on version 7 that it is unusable for us.  We are seeing it on both Android and iOS devices of many different kinds.  We are currently still trying to make version 6.0 work, but are running into other things that can't be solved on version 6.  


Thanks,

Heather 

Hi Mark,



I would not expect the SDK 7.1 release to be inferior in terms of tracking performance or computational performance. We do test our releases quite extensively and I think this should definitely have showed up. I'm also not aware of any similar reports.


May I ask which device(s) you are experiencing this on?


Also, have you tried the SDK 7.0 release by any chance? Or did you go straight to 7.1 from 6?



- Daniel

It was fast and easy to recognize images with wikitude SDK 6, it became slowly and hard when we change to SDK 7.1. Our recognition is significantly worse on version 7.1 than it is on version 6. So much worse that version 7 isn't usable for us. Any ideas what the difference could be between version 6 and version 7 with the recognition? Is there any way to get version 7 to recognize as well as version 6?

 

 Our developing situation : Android + javascript + wikitude SDK 7.1 + Android studio (2.3.3) + wtc file (version 5).

 

        Mark Cai

wtc
(30.6 KB)
js
Login or Signup to post a comment