| --- |
| tags: |
| - smart-manufacturing |
| - sft |
| - industrial |
| - vision |
| license: other |
| pretty_name: "183" |
| extra_gated_fields: |
| Name: text |
| Affiliation: text |
| Intended use: text |
| extra_gated_prompt: >- |
| This dataset is released for **research use**. Access is reviewed and granted |
| **manually** by the maintainers. Please state your name, affiliation, and intended use. |
| --- |
| |
| # 183 |
|
|
| PCB defect detection (6 classes; VOC bbox). Category **B**, task **T-B2**, in the unified Smart-Manufacturing SFT schema. |
|
|
| > The repository name is an internal task code. See **Provenance** below for the underlying dataset. |
|
|
| ## Records |
|
|
| **693** records (train=693). |
|
|
| ## Unified SFT schema |
|
|
| | field | type | meaning | |
| |---|---|---| |
| | `query` | str | the question / instruction (model input) | |
| | `image` | Image | the input image (bytes embedded) | |
| | `annot` | str | the answer — for this dataset: one `class,[x, y, w, h]` line per defect bounding box (COCO x/y/width/height in pixels; converted from the source Pascal-VOC corner boxes). The 6 PCB-defect classes are a closed set given in the query; full boxes + image size are in `metadata.objects`. Detection task — no mask column — see **Task & split** below | |
| | `reasoning` | null | no native CoT in these datasets | |
| | `cate` | "B" | SFT category | |
| | `task` | "T-xx" | unified task id | |
| | `metadata` | str (JSON) | split, provenance, `image_path`, `image_sha256` (dedup key) | |
| | `mask` | Image \| null | *(T-B1/T-B2 only)* the pixel ground-truth mask, bytes embedded | |
| | `masks` | list[Image] | *(D21 only)* multi-region masks | |
|
|
| ## Task & split |
|
|
| **What this is.** HRIPCB — the Peking University PCB Defect Dataset (Ding et al.): 693 high-resolution |
| printed-circuit-board images organized by defect type into **6 classes** (missing_hole, mouse_bite, open_circuit, |
| short, spur, spurious_copper), each annotated with Pascal-VOC bounding boxes (2,953 boxes total; each image holds |
| several boxes of its one defect type). Every image contains defects — no defect-free images. |
|
|
| **Task.** Object **detection**: localize and classify every PCB defect. `query` (our template) names the closed |
| set of 6 classes and asks for one `class,[x, y, w, h]` line per box (top-left x, y + width, height, in pixels). |
| `annot` is that — source VOC corner boxes converted to COCO `[x,y,w,h]`. **There is no mask** — localization is |
| the bounding box. Full boxes + image size are in `metadata.objects`; `metadata.category` records the source |
| folder (the image's defect type). |
|
|
| **Split.** No upstream train/test split -> single `train` (693). |
|
|
| ## Provenance |
|
|
| Underlying dataset: **HRIPCB**. Upstream license: **other (research use; Peking Univ. PCB Defect Dataset)** (this card is `license: other`; respect the upstream terms). Converted read-only from the raw source into the unified schema; conversion script: `183/convert_d83.py`, published with `publish/push_to_hf.py`, both in [`AI4Manufacturing/forge_model`](https://github.com/AI4Manufacturing/forge_model). |
|
|
| ## Overlap / de-duplication (§8) |
|
|
| None notable. Each record carries `metadata.image_sha256` so overlapping images can be kept entirely on one side of a train/eval split. |
|
|
