Recyclable Garbage Classification (MMoRWaD)

Code for The Multi Modal Recyclable Waste Management Dataset workflow using synchronized RGB, depth, video, and impact audio.

Temporary / complementary repository: This code repository complements the dataset repository and the related publication material. It will be updated and further cleaned when I become available.

What’s in this repo

• Dataset pipeline: tools for dataset creation/refinement/annotation/validation (scripts + helpers)
• Classification pipeline: experiments, training utilities, and inference scripts for multimodal MSW classification

What’s NOT in this repo

To keep this GitHub repository lightweight and reproducible:

• The dataset itself is not included (see the Zenodo link below).
• Model weights/checkpoints are not included (*.keras, *.h5, *.pt, *.pth, *.ckpt, etc.).
• Local data folders such as dataset/ and predict_data/ are ignored.

Links / identifiers

• Paper DOI: [DOI_PLACEHOLDER_TO_BE_ADDED_AFTER_PUBLICATION]
• Zenodo dataset link: https://zenodo.org/records/18364275

Repository layout (high level)

• DatasetPipeline/ — dataset curation, annotation tools, validators, utilities
• Capture/ — capture scripts and helper utilities
• DataExploration/ — exploratory notebooks (image/audio analysis)
• NeuralNetwork/ — training code + environment files
• OtherClassifiers/ — additional classifier experiments
• Prediction/ — inference / prediction scripts

Setup (recommended)

Environment references are provided in:

• NeuralNetwork/nnenv.yml
• NeuralNetwork/gpu_nnenv.yml

Example (Conda):

1. Create environment (CPU example):

   conda env create -f NeuralNetwork/nnenv.yml
   conda activate nnenv

2. Run scripts from the relevant submodule (DatasetPipeline/, NeuralNetwork/, Prediction/).

Notes:

• Some scripts expect local paths to the dataset; those inputs are intended to be provided via CLI args and/or config variables (dataset not shipped here).
• If you encounter missing-package errors, use the environment YAMLs as the source of truth.

Citation

Please cite the paper and the dataset/repository resources:

• Paper DOI: [DOI_PLACEHOLDER_TO_BE_ADDED_AFTER_PUBLICATION]
• Dataset (Zenodo): https://zenodo.org/records/18364275

Notes / acknowledgments

• This is a temporary complementary code repository to the dataset repository and will be updated.
• Final code retouching/restructuring and repository preparation were completed with assistance from GitHub Copilot (GPT-5.2).