Waste Detection on the browser using TFLite model

As an initiative to solve for environment, this project is an implementation of detecting various categories of waste in real-time by deploying a TF Lite model directly on the browser using the TFJS-TFLite API.

Note: TF Lite Model Maker is now obsolete and is replaced by MediaPipe Model Maker.

Implementation details:

The object detection model was trained on a custom dataset of different categories of waste – open litter, plastic waste, biodegradable waste, medical waste, and overflowing dustbin.

Steps to run:

1. Clone the repository on your local machine.

2. For static detection, navigate to the directory cd Static Detection; for real-time detection, navigate to the directory cd Real-time Detection.

3. Open your terminal/command prompt and enter the command py -m http.server (if you have Python installed) to create a local server.

4. Open your web browser, and go to localhost:8000.

Static Detection:

Real-time Detection:

• Try a LIVE demo on CodeSandbox.
• Read the Medium blog for technical details.
• Watch the full demonstration on YouTube.
• The project is featured on Google Dev Library.

References:

• TF Lite Model Maker tutorial: To train a custom object detection TF Lite model.
• TFJS-TFLite Task API: To load the TF Lite model directly on the browser and do inferencing.