TFLite Mobile Generic Object Localizer

Python TFLite scripts for detecting objects of any class in an image without knowing their label.

Image taken from the OpenCV AI Kit - Lite, make sure to check it out: https://www.kickstarter.com/projects/opencv/opencv-ai-kit-oak-depth-camera-4k-cv-edge-object-detection

❗ ⚠️ The object detector works better with images with few objects and it starts to fail in more complex scenes. The model is suitable for automatically labelling objects for custom object detection models.

Requirements

• OpenCV, imread-from-url and tensorflow or tflite_runtime. Also, pafy and youtube-dl are required for youtube video inference.

Installation

pip install -r requirements.txt
pip install pafy youtube-dl

For the tflite runtime, you can either use tensorflow(make sure it is version 2.6.0 or above) pip install tensorflow==2.6.0 or the TensorFlow Runtime binary

tflite model

The original models was taken from Tensorflow Hub, download it, and place it in the models folder.

Use the following script to download the model:

python download_model.py

Tensorflow inference

Use this other repository for detecting object in Tensorflow (recommended for computer with GPU): https://github.com/ibaiGorordo/Tensorflow-Mobile-Generic-Object-Localizer

Examples

• Image inference:

python imageObjectDetection.py 

• Webcam inference:

python webcamObjectDetection.py 

• Video inference:

python videoObjectDetection.py

Inference Examples

Original video by Animist: https://youtu.be/uKyoV0uG9rQ

Cabybara detection

Original image: https://commons.wikimedia.org/wiki/File:Capybara_portrait.jpg

Coin detection

Original image: https://commons.wikimedia.org/wiki/File:Japanese_Coins.jpg

Shoe detection

Original image: https://commons.wikimedia.org/wiki/File:Japanese_Coins.jpg

Spaceship detection

Original image: https://en.wikipedia.org/wiki/Spacecraft#/media/File:SpaceX_Crew_Dragon_(More_cropped).jpg

Window detection

Original image: https://commons.wikimedia.org/wiki/File:Window_-_Paddington_-_London.JPG

And many more

References:

• Original model: https://tfhub.dev/google/lite-model/object_detection/mobile_object_localizer_v1/1/default/1