tensorflutter

Internship example of TensorFlow Lite object detection in Flutter. One screen shows eight targets. Tapping one opens the camera and speaks when that class is seen.

Published September 2020. The last commit, in March 2021, only renamed the project.

Plugins named in pubspec.yaml:

• tflite ^1.1.1
• camera ^0.5.8+5
• flutter_tts ^1.2.7
• flutter_staggered_grid_view ^0.3.2

pubspec.lock is gitignored, so those are ranges, not a frozen resolution.

Notes from 18 October 2026

Reviewed again on this date. These points describe the code as it stands; they are not part of the original readme.

• Snapshot of the 2020 tree. It is not being ported to current Flutter.
• Dart constraint is >=2.7.0 <3.0.0. flutter pub get fails on Dart 3.8 (Flutter 3.32) and on any null-safe Dart SDK. .metadata records stable revision b041144f.
• Every tile calls loadModel(), and that function always loads assets/ssd_mobilenet.tflite. Tiny YOLOv2 (43 MB), MobileNet (16 MB), and PoseNet (4.8 MB) are in the repo and have code paths, but the screen never selects them.
• The phone tile looks for the label phone. assets/ssd_mobilenet.txt names that class cell phone, so that tile never draws a box or speaks. The other seven names are in the label file.
• The grid images are hotlinked from freedesignfile.com. They are not stored in this repo. If those URLs go away, the tiles have no picture.
• Speech is a one-second timer. Leaving the page does not cancel it. If a detection call fails, isDetecting stays true and the camera stream stops feeding the model.
• iOS Info.plist has no camera usage description. The Android application id is com.soft4earth.tensorfirat.tensorfirat. The release build is signed with the debug key.
• .DS_Store, .iml, and .idea are committed. .idea contains an old local Flutter path under /Users/meg/development/flutter. No API key or token is in the tree or the git history.
• The widget test is still the Flutter counter sample. There are no GitHub issues on this repository.