boscochanam/circuit-component-detector

Circuit Component Detector (YOLO26M-OBB, 16 classes)

Oriented-bounding-box detector for hand-drawn electronic circuit components. It is Stage 1 of the pipeline in "From Hand-Drawn Schematics to SPICE Netlists" — it localizes and orients components so a downstream occlusion + graph-join stage can recover electrical connectivity.

• Architecture: YOLO26M-OBB (Ultralytics), fine-tuned.
• Task: oriented bounding-box detection.
• Classes: 16, merged down from CGHD-1152's 61 (variants the netlist does not distinguish are collapsed; 30 rare device types fold into a single other class).
• Reported performance: mAP@0.5 = 89.0% on 468 held-out scans.

Intended use

Detecting and orienting components in scanned/photographed hand-drawn schematics as the first stage of schematic-to-netlist digitization. The model recovers component location, orientation, and coarse class — not device values. It is trained on a single hand-drawn corpus; generalization to printed schematics or other drawing styles is unvalidated.

Training data & attribution

Trained on CGHD-1152, A Public Ground-Truth Dataset for Handwritten Circuit Diagram Images by Felix Thoma, Johannes Bayer, Yakun Li, and Andreas Dengel (DFKI), licensed CC BY 4.0 and archived at Zenodo: https://doi.org/10.5281/zenodo.6385814.

License

The base weights are Ultralytics YOLO26, released under AGPL-3.0. This fine-tuned derivative is therefore distributed under AGPL-3.0. If you need a non-AGPL license for the weights, obtain an Ultralytics Enterprise License. (The surrounding pipeline code is MIT and the CGHD-derived annotations are CC BY 4.0 — see the repository — but those licenses do not extend to these weights.)

Links

• Code: https://github.com/boscochanam/circuit-digitization
• Archived release (Zenodo): https://doi.org/10.5281/zenodo.21274158 (all versions) · https://doi.org/10.5281/zenodo.21274159 (v1.0.1)

Citation

If you use this model, please cite the paper and the software archive:

@article{chanam2026circuitdigitization,
  title   = {From Hand-Drawn Schematics to SPICE Netlists: A Deterministic
             Pipeline with Endpoint-Graph Wire Joining and a Human-Verified
             Connectivity Benchmark},
  author  = {Chanam, Bosco and Dcosta, Chris and Talupuri, Pranavesh Kumar and
             Chiwhane, Shwetambari and Singh, Ashay Kumar and Das, Arghadeep},
  journal = {IEEE Access},
  year    = {2026},
  note    = {Under review}
}

@software{chanam2026circuitdigitization_sw,
  title     = {Circuit Digitization: a deterministic hand-drawn-schematic-to-SPICE
               pipeline with an endpoint-graph wire join and a human-verified
               connectivity benchmark},
  author    = {Chanam, Bosco and Dcosta, Chris and Talupuri, Pranavesh Kumar and
               Chiwhane, Shwetambari and Singh, Ashay Kumar and Das, Arghadeep},
  year      = {2026},
  version   = {1.0.1},
  doi       = {10.5281/zenodo.21274158},
  url       = {https://github.com/boscochanam/circuit-digitization}
}

And the training dataset:

@inproceedings{thoma2021cghd,
  title     = {A Public Ground-Truth Dataset for Handwritten Circuit Diagram Images},
  author    = {Thoma, Felix and Bayer, Johannes and Li, Yakun and Dengel, Andreas},
  booktitle = {Proc. Int. Conf. Document Analysis and Recognition (ICDAR)},
  pages     = {20--27},
  year      = {2021},
  doi       = {10.1007/978-3-030-86198-8_2}
}