| --- |
| license: other |
| license_name: research-use-mixed-sources |
| license_link: LICENSE.md |
| task_categories: |
| - object-detection |
| tags: |
| - pcb |
| - component-detection |
| - object-detection |
| - computer-vision |
| - yolo |
| - electronics |
| - printed-circuit-board |
| - pcb-components |
| - electronic-components |
| - manufacturing |
| - industrial-inspection |
| pretty_name: PCB Component Detection Dataset (High-Resolution) |
| size_categories: |
| - 1K<n<10K |
| --- |
| |
| # PCB Component Detection Dataset (High-Resolution) |
|
|
| A YOLO-format object detection dataset for locating and classifying **electronic components on a printed circuit board** — resistors, capacitors, ICs, diodes, connectors, and more — from full high-resolution board photographs. |
|
|
| This dataset powers [`ParsaVictor/pcb-component-detection-yolo26`](https://huggingface.co/ParsaVictor/pcb-component-detection-yolo26), part of the [PCB Inspection Suite](https://huggingface.co/ParsaVictor/pcb-inspection-suite). |
|
|
| ## Why high-resolution matters |
|
|
| A naive 640×640 export (the Roboflow default) shrinks small components like resistors down to a handful of pixels, which caps detector accuracy hard. This dataset was rebuilt from **original full-resolution board photographs** (median part size ~19px at the chosen tiling, vs ~6px in the original low-res export), with labels carefully re-mapped back onto the originals. |
|
|
| ## Classes (13) |
|
|
| | Class | # instances | |
| |---|---| |
| | Resistor | 34,470 | |
| | Capacitor | 33,923 | |
| | IC | 12,015 | |
| | Electrolytic Capacitor | 10,508 | |
| | Transistor | 4,049 | |
| | Diode | 3,904 | |
| | Inductor | 3,402 | |
| | Connector | 2,867 | |
| | Marker | 2,465 | |
| | Other | 1,680 | |
| | LED | 444 | |
| | Switch/Button | 287 | |
| | Oscillator | 268 | |
|
|
| ## Splits |
|
|
| | Split | Images | |
| |---|---| |
| | train | 3,005 | |
| | val | 321 | |
| | test | 292 | |
|
|
| Images are tiled crops from full-resolution board photographs (1024px tiles, 25% overlap) so components stay detectable at training resolution while the full board can still be reconstructed at inference time. |
|
|
| ## Format |
|
|
| Standard YOLO format: |
| ``` |
| images/{train,val,test}/*.jpg |
| labels/{train,val,test}/*.txt # class_id x_center y_center width height (normalized) |
| data.yaml |
| ``` |
|
|
| A simplified **binary "any component" variant** (single class, same images) is also used internally as a first detection stage before fine-grained classification; see the suite overview for the two-stage pipeline. |
|
|
| ## Quick start |
|
|
| ```python |
| from ultralytics import YOLO |
| |
| model = YOLO("yolo26s.pt") |
| model.train(data="data.yaml", imgsz=1024, epochs=80) |
| ``` |
|
|
| ```python |
| from huggingface_hub import snapshot_download |
| path = snapshot_download(repo_id="ParsaVictor/pcb-component-detection", repo_type="dataset") |
| ``` |
|
|
| ## Related |
|
|
| - Sibling dataset: [pcb-bare-board-defect-detection](https://huggingface.co/datasets/ParsaVictor/pcb-bare-board-defect-detection) |
| - Sibling dataset: [pcb-solder-defect-detection](https://huggingface.co/datasets/ParsaVictor/pcb-solder-defect-detection) |
| - Full suite overview: [pcb-inspection-suite](https://huggingface.co/ParsaVictor/pcb-inspection-suite) |
|
|
| ## License / attribution |
|
|
| Released for **research and development use**. Built from manually re-annotated, full-resolution board photographs. Contact the author before any commercial redistribution. See `LICENSE.md`. |
|
|
