Supercharging Android Apps with Tensorflow
summary
Shows a working Android app that embeds TensorFlow C++ models via JNI and Bazel, converts camera bitmaps to RGB tensors, and runs real‑time image classification entirely on‑device (airplane mode). Model is the Inception-derived bundle that the builder reports takes ~2 hours to build and produces about a 60 MB apk that recognizes ~10,000 objects. Requires Android 5+, NDK and Bazel, and demo results show usable confidence scores though uncommon objects drop accuracy; also mentions person detection and on‑device style transfer examples.
AI-written summary · about
chapters (12)
- 0:00Start
- 2:21Outline
- 3:03Machine Learning is Not
- 7:27Machine Learning Workflow
- 8:42Types of machine learning
- 12:16Deep learning: Neural Nets
- 16:11A Tensorflow Program
- 17:16A little of TensorFlow
- 18:05Why TensorFlow For Android
- 19:07Android TensorFlow example
- 23:12App Structure
- 25:37Build Run
related talks
19mOn-device machine learning: TensorFlow on Android
40mApplied TensorFlow in Android apps
43mTensorFlow for Mobile Developers
33mWho’s Afraid of Machine Learning? First Steps with TensorFlow for Android
29mFrom TensorFlow to ML Kit: power your Android application with ML
41mMaking Android Apps with Intelligence
30mTensorFlow for Mobile Heroes
42mCreating Smarter Applications