google-ai-edge / LiteRT

LiteRT is the new name for TensorFlow Lite (TFLite). While the name is new, it's still the same trusted, high-performance runtime for on-device AI, now with an expanded vision.
https://ai.google.dev/edge/litert
Apache License 2.0
169 stars 13 forks source link

Samsung S21: Noisy AI Inference Results with App Animations in PlayService/GPU Mode #170

Open gaikwadrahul8 opened 3 days ago

gaikwadrahul8 commented 3 days ago

System information

I followed the tutorial from

https://www.tensorflow.org/lite/android/play_services#java_6.

Everything seemed fine on almost all devices. However, our test team recently attempted to test our app on the Samsung S21 and discovered that the AI inference results are noisy when there are animations on the app's GUI. This issue only occurs in the PlayService/GPU scenario. If I switch to standalone mode, everything works well. We also found that setting "android:hardwareAccelerated" to false in the AndroidManifest.xml can produce stable results, but it's obviously not a reasonable solution. Has anyone else experienced a similar issue? I would appreciate any advice or suggestions.

gaikwadrahul8 commented 3 days ago

This issue originally reported by @JeremyJian has been moved to this dedicated repository for LiteRT to enhance issue tracking and prioritization. To ensure continuity, we have created this new issue on your behalf.

We appreciate your understanding and look forward to your continued involvement.