Skip to content

v3

Choose a tag to compare

View all tags
@tit-exe tit-exe released this 04 Jun 13:57
· 7 commits to main since this release
release
04d78fa

OrangIdentifier Android - V3.0

Release of the 3rd major version of the OrangIdentifier Android companion app. This offline mobile application allows field rangers to photograph Bornean orangutans and identify them in real-time, completely without internet access.

Historical context: V1 used a ResNet closed-set classifier, V2 moved to ResNet embeddings. This V3 release introduces a major architectural upgrade with MegaDescriptor.

🌟 Key Features

  • 100% Offline Inference: Runs entirely on-device using optimized TensorFlow Lite models.
  • V3 Pipeline (MegaDescriptor): Uses the state-of-the-art V3 pipeline with a YOLO v2 face detector and a MegaDescriptor-T backbone for high-accuracy open-set recognition (reliably rejects unknown individuals).
  • Instant Onboarding: Add new individuals directly from the app in less than a minute. The embedding is immediately saved to the local gallery without any model retraining.
  • Scan History: A local Room database keeps a log of all past identifications.
  • Real-time Correction: Correct wrong predictions on the fly to continuously improve an individual's prototype vector.

📥 Installation Instructions

  1. Download the orangs_android_app_v3.apk attached to this release under Assets.
  2. Open the APK file on your Android device to install it. (You may need to enable "Install from unknown sources" in your Android settings).

⚠️ Note for Developers

This APK is ready to use. However, if you are building the project from source, remember that the heavy backbone model (megadesc_T_arcface_backbone.tflite, 112MB) is not included in the source code due to GitHub file size limits. You must download it manually from the HuggingFace repository and place it in the app/src/main/assets/ directory before building.

Assets 3
Loading