Dataset

Browse current records at app.cml-relab.org. The app shows live, evolving records. A curated dataset release is a reviewed, versioned snapshot with defined scope, metadata, and licensing, suitable for citation and reuse.

What a release will contain

Expected data elements:

• product metadata
• component hierarchies
• images and possibly video-linked records
• material and category annotations
• measurements and observational notes

Expected uses:

• computer vision and image-based classification tasks
• circular economy and design-for-disassembly studies
• material composition analysis
• comparative studies across product families or brands
• industrial ecology and LCA-oriented work that depends on primary product data
• linking with other open industrial-ecology data infrastructures

Access

Explore current records in the app or through the API; see the API reference overview or the public API reference.

The planned release license is CC BY 4.0. No release has been published yet. The dataset codebook documents every file and column a release carries.

Dublin Core metadata

Following the Dublin Core specifications:

Element	Value
Title	Disassembly dataset of durable goods for circular economy and computer vision applications
Creators	van Lierde, Simon: https://orcid.org/0009-0006-6953-909X
Creators	Donati, Franco: https://orcid.org/0000-0002-8287-2413
Publisher	Leiden University. Institute of Environmental Sciences (CML)
Subjects	Computer vision, Circular economy, Remanufacturing, Life cycle assessment, Durable goods
Description	Data collection platform for disassembled power tool images and metadata supporting computer vision tasks for life cycle assessments.
Date	2025-03/2026-08 (collection ongoing)
Types	Dataset, Image, Software
Formats	text/csv, image/jpeg, application/x-python, text/markdown
Identifier	https://github.com/CMLPlatform/relab
Language	en-US
Coverage	Products: Power tools; Time: 2025-03–2026-08, ongoing; Geographic location: NL
Rights	https://creativecommons.org/licenses/by/4.0/

Building and depositing a release

(For maintainers.) The pipeline lives in backend/scripts/ and runs from backend/:

1. just release-build --inventory: a dry run that writes nothing and reports what a release would contain. Only consenting accounts are included: owners who accepted the contributor terms at the version that grants the publication licence. Everyone else is excluded by the no-terms-acceptance rule and listed in excluded-records.csv. Accounts created before acceptance was tracked hold no grant until they answer the in-app prompt, and declining is a real outcome.
2. just release-build --out dist/dataset-vX.Y --software-doi 10.5281/zenodo.NNNNNNN: builds the release directory. The software DOI is the version DOI of the Relab release the data was exported from, listed on the concept record; the provenance link must name one frozen release. The verification pass at the end fails the build rather than warning.
3. Review dist/dataset-vX.Y/review/ by hand, including excluded-records.csv and the rule that excluded each record. This directory is not part of the published archive.
4. The pseudonymisation salt comes from RELAB_PSEUDONYM_SALT (preferred), --pseudonym-salt, or secrets/<env>/dataset_pseudonym_salt. Keep the same salt for every release so owner pseudonyms stay stable across versions.
5. uv run python -m scripts.zenodo_deposit --dir dist/dataset-vX.Y creates or versions a Zenodo draft. Add --deposition <id> to resume uploading into an existing draft. The token is read from secrets/<env>/zenodo_token (or ZENODO_TOKEN for a one-off run).
6. Inspect the draft in the Zenodo web UI, then publish with --deposition <id> --publish. A published record can be tombstoned but never withdrawn or edited, so the command prompts before it sends anything.

Full flags: just release-build --help and uv run python -m scripts.zenodo_deposit --help.

Contact

For questions about using the dataset or platform, contact [email protected].