undefined value

[shahinyan11]

Below you can see a part of my code where the call detectObjects(frame, frameProcessorConfig) always returns undefined

I also looked at the logs in Android Studio and found this

Is there someone, who familiar with this error and managed to use the package without mentioned error ?

[hirendudhat03]

@shahinyan11 your model include metadata? because I have same issue because my model not including metadata.

[Noahkoole]

Have you been able to fix this issue?

[arpkirill]

Have you been able to fix this issue?

no

[Noahkoole]

@arpkirill Your logs in android studio explains the issue. You have 2 different kinds of tflite models. 1 with metadata including labels and stuff and 1 without. you can only use the tflite with metadata. if you take an example model from tflite website it works fine. I assume it uses MLKit or something underneath because MLKit also only works with metadata.

[arpkirill]

@arpkirill Your logs in android studio explains the issue. You have 2 different kinds of tflite models. 1 with metadata including labels and stuff and 1 without. you can only use the tflite with metadata. if you take an example model from tflite website it works fine. I assume it uses MLKit or something underneath because MLKit also only works with metadata.

Tell me a fully working tflite file so that I can test

[arpkirill]

@arpkirill Your logs in android studio explains the issue. You have 2 different kinds of tflite models. 1 with metadata including labels and stuff and 1 without. you can only use the tflite with metadata. if you take an example model from tflite website it works fine. I assume it uses MLKit or something underneath because MLKit also only works with metadata.

https://github.com/arpkirill/test-camera-object-detector

this is my demo project, which has a problem, I will be glad to hear your comments

[jejolare]

I have the same issue, tried several models. Btw, @arpkirill, your repository is private, we can't open

[jejolare]

@Noahkoole , could you provide any example of fully working model please 🙏?

[arpkirill]

I have the same issue, tried several models. Btw, @arpkirill, your repository is private, we can't open

I made it public, the only thing is to manually give permission to the phone camera

[arpkirill]

I have the same issue, tried several models. Btw, @arpkirill, your repository is private, we can't open

do you have any idea how to run it?

[jejolare]

arpkirill

idk, I can't run even their example. I have same issue, and have no idea what to do. I don't think that developer will help us soon:

[jejolare]

or maybe someone could provide us with working tflite

[arpkirill]

arpkirill

idk, I can't run even their example. I have same issue, and have no idea what to do. I don't think that developer will help us soon:

In the demo project, you need to update the dependencies in Gradle, I don't remember which ones exactly. Did you launch my project?

[jaroslawkrol]

@arpkirill @jejolare As @Noahkoole and @hirendudhat03 mentioned above your model has to contains metadata. Remember about it if you wants to create your own TFLite model. It's not my library's requirement, it's just a TensorFlow Lite Vision requirement used in the background.

If you want to test your code with existing correct model, just use TFHub as I mentioned in docs. To do it:

1. Go to TFHub.dev
2. In filters select Problem domains > Images > Object Detection
   
3. Find one of models matched to your preferences and contains metadata included f.e. ssd_mobilenet_v1
   
4. In section Model formats click on tab with (metadata) if there is more than one format
   
5. Download it!

Sorry guys, for so late response but as I mentioned in one of the threads my current project and private life consumes me totally. I also have my doubts that it still makes sense to maintain the library, because the new version of Vision Camera is going to bring big changes and possibility to use TFLite models just by drag & drop. But I need to talk with Marc about it to be sure, what is the current direction of Vision Camera library.

@arpkirill Please, test your code with this model: https://tfhub.dev/tensorflow/lite-model/ssd_mobilenet_v1/1/metadata/2 and let me know about results, to know i can close this ticket. If not, we will try to resolve it together.

[arpkirill]

@arpkirill @jejolare As @Noahkoole and @hirendudhat03 mentioned above your model has to contains metadata. Remember about it if you wants to create your own TFLite model. It's not my library's requirement, it's just a TensorFlow Lite Vision requirement used in the background.

If you want to test your code on existing correct one, just use TFHub as I mentioned in docs. To do it:

1. Go to TFHub.dev
2. In filters select Problem domains > Images > Object Detection
   
3. Find one of models matched to your preferences and contains metadata included f.e. ssd_mobilenet_v1
   
4. In section Model formats click on tab with (metadata) if there is more than one format
   
5. Download it!

Sorry guys, for so late response but as I mentioned in one of the threads my current project and private life consumes me totally. I also have my doubts that it still makes sense to maintain the library, because the new version of Vision Camera is going to bring big changes and possibility to use TFLite models just by drag & drop. But I need to talk with Marc about it to be sure, what is the current direction of Vision Camera library.

@arpkirill Please, test your code with this model: https://tfhub.dev/tensorflow/lite-model/ssd_mobilenet_v1/1/metadata/2 and let me know about results, to know i can close this ticket. If not, we will try to resolve it together.

Everything works fine with this model, thank you very much for your help! I saw version 3, it looks cool, but as I understand it, it is still in the beta version.

[jaroslawkrol]

@arpkirill Thanks for quick reply! BTW. the prettiest cat i've ever seen 🤣

[ElicaInc]

Super! Great tutorial!