| --- |
| tags: |
| - smart-manufacturing |
| - sft |
| - industrial |
| - vision |
| license: other |
| pretty_name: "184" |
| 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. |
| --- |
| |
| # 184 |
|
|
| Electrical-commutator surface-defect binary detection (segmentation GT). Category **B**, task **T-B1**, in the unified Smart-Manufacturing SFT schema. |
|
|
| > The repository name is an internal task code. See **Provenance** below for the underlying dataset. |
|
|
| ## Records |
|
|
| **399** records (train=399). Pixel masks are embedded as a `mask` image column. |
|
|
| ## 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: the plain-text image-level label `good` or `anomalous` (binary; no defect types — the label is derived from the pixel mask). The binary segmentation `mask` is deferred localization GT, with seg info (`mask_path`, `defect_area_fraction`) in `metadata` — see **Task, mask & 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, mask & split |
|
|
| **What this is.** KolektorSDD (Tabernik et al., *"Segmentation-Based Deep-Learning Approach for Surface-Defect |
| Detection"*, J. Intelligent Manufacturing 2020) — 399 grayscale images of electrical-commutator surfaces from |
| 50 items (8 sections each); 52 defective, 347 defect-free. Each image ships a pixel-level binary defect mask. |
|
|
| **Task & label.** Surface-defect detection: image-level binary (defect vs OK) + pixel-level segmentation. The |
| source has no good/defect folders — the image-level label is **derived from the mask** (any nonzero pixel -> |
| anomalous). `query` (our template) asks only whether the surface is **good** or **anomalous**; `annot` is the |
| plain-text answer `good` or `anomalous`. **The query does not ask for a mask.** |
|
|
| **Segmentation (deferred GT).** The binary segmentation mask is kept in the `mask` column as localization ground |
| truth (anomalous images only; good images have `mask`=null). Per-image seg info is in `metadata`: `mask_path` and |
| `defect_area_fraction`. A text-output model cannot emit a pixel mask, so segmentation is deferred. |
|
|
| **Split.** No train/test split in the source (the paper uses 3-fold cross-validation) -> single `train` (399 |
| images: 52 anomalous + 347 good). |
|
|
| ## Provenance |
|
|
| Underlying dataset: **KolektorSDD**. Upstream license: **CC BY-NC-SA 4.0** (this card is `license: other`; respect the upstream terms). Converted read-only from the raw source into the unified schema; conversion script: `184/convert_d84.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. |
|
|
