Object detection on smartphone [WIP]
App to demonstrate object detection on smartphones using Flutter & Tensorflow Lite frameworks.
Constrains
- min Sdk version 21 => Works with Android 5 & above
- Tested on Smartphone with Android 9 -- Works smoothly
- Tested on Smartphone with Android 6 -- Has issues
Tech Stack
- Flutter
- Tensorflow Lite
- COCO SSD MobileNet Deep learning model
TODOs
- Landscape mode
- Inffered box and label on saved image
- label at location of box
- measure performance
- exception handling
- Fix issues occuring on android 6
How to use?
- Open App -- Camera preview will be shown with detected objects in the bounding boxes with their class mentioned above
- Press capture button to click image with objects in bounding boxes with their class mentioned above
- Press the same button to go back and save that image
Screenshots
- Near-real-time Object inference
- Image saved to Gallery
How to run locally?
Resources/References
- https://www.tensorflow.org/lite/models/object_detection/overview#get_started
- https://api.flutter.dev/flutter/widgets/CustomPaint-class.html
- https://pub.dev/packages/tflite
- https://pub.dev/packages/camera
- https://pub.dev/packages/image
- https://pub.dev/packages/permission_handler
- https://medium.com/swlh/build-a-dog-camera-using-flutter-and-tensorflow-312022b716c6
- https://pub.dev/packages/image_save