This dataset is released for research use. Access is reviewed and granted manually by the maintainers. Please state your name, affiliation, and intended use.

Roles

Roles: canon repo — annot is the source label, kept machine-parseable as the gold for verification and reward parsing; there is no reasoning column and this repo is not itself a training view. Derived repos (-annotated, -grounding, -region, -mcq) each state their own regime on their own card. Geometry for every record lives in metadata.geometry (below).

208

Power-line cable anomaly detection & classification (7 anomaly types, expert severity grades; COCO 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

4,798 records (train=4798).

Unified SFT schema

field	type	meaning
`query`	str	the question / instruction (model input)
`image`	Image	the input image (bytes embedded); for multi-image rows, a preview of the first view
`images`	list[Image]	(multi-image rows) all input views / modalities for the row, bytes embedded
`annot`	str	the answer — for this dataset: one line per anomaly, `class,[x, y, width, height]` (plain text). Coordinates are NATIVE-pixel COCO xywh at 1920x1080 — top-left origin, NOT xyxy, NOT normalized; convert per student model (see `common/box_convert.py`). Defect-free images answer `none` — they are 45% of the set and are kept on purpose. The expert severity grade and a cross-frame `anomaly_id` ride in `metadata.objects[*].attributes` and are NOT asked
`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]	(multi-image T-B1 / D21) per-view masks aligned with `images` (None where a view has no defect), or multi-region masks

Task, severity, split & counts

What this is. CableInspect-AD (Arodi, Luck, Bedwani, Zaimi, Li, Pouliot, Beaudry, Marceau Caron, "CableInspect-AD: An Expert-Annotated Anomaly Detection Dataset", NeurIPS 2024 Datasets & Benchmarks), released by Hydro-Quebec under CC BY-NC-SA 4.0. 4,798 high-resolution 1920x1080 images of three power-line cables photographed along their length, expert-annotated in COCO format with 7 anomaly categories: bent strand, broken strand, crushed, deposit, long scratch, spaced strand, welded strand.

This is field maintenance inspection — a power-transmission asset in service, not a part on a production line — and it is the only dataset of that kind in this corpus.

Task & answer. Detection + classification, the same shape as D23. query asks for every anomaly with its class and bounding box; annot is one line per anomaly, class,[x, y, width, height].

Defect-free images are kept, and answer none. 2,159 of the 4,798 images carry no annotation. A pure detection set usually drops those — but here they are the normal class of an anomaly benchmark, and dropping them would leave a corpus in which every image contains a defect. none is used rather than an empty string so a blank answer cannot be confused with a truncated one.

⚠ No official split — folds must be CABLE-WISE. The authors evaluate by cross-validation over the three cables, so every record lands in a single train split carrying metadata.subset (cable_1 = 2,054 images, cable_2 = 1,567, cable_3 = 1,177); the capture pass is the numbered directory in metadata.image_path. Group folds by cable. Splitting by image puts near-duplicate frames of the same cable segment on both sides — the same grouped- split rule this corpus applies to bearings and scenes.

Severity is preserved, not asked. Every one of the 3,864 annotations carries an expert gradation grade and a stable anomaly_id that identifies the same physical defect across frames; both sit in metadata.objects[*].attributes, with a flattened metadata.severity_grades / metadata.anomaly_ids per record. Observed grades: light (1,755), important (758), extracted (339), partial (212), deep (115), complete (102), superficial (80). The vocabulary is category-dependent and the release documents no ordering across categories, so severity is preserved as ground truth rather than turned into a target. anomaly_id also lets you group the frames that show one physical anomaly.

⚠ Annotation count differs from the paper. The three bbox JSONs contain 3,864 annotations and the three _seg.json files a further 976 (segmentation, covering a subset of the same anomalies); the paper reports "6,023 annotated anomalies". We could not reconcile that from the release, so this repo carries exactly what cable_{1,2,3}.json holds — verified: 3,864 objects across 2,639 annotated images — and states the discrepancy rather than repeating the paper's number. The _seg.json polygons are not merged in.

Lazy-baseline floor. 2,159 of 4,798 images (45.0%) are defect-free, so always answering none scores 45.0% on the presence question. Detection quality must be reported as IoU-matched precision/recall, not as answer accuracy.

Query text — pooled paraphrases (v2)

Every record's query is drawn from common/vision_query_pools.json[208/orig], a pool of 37 gate-verified paraphrases of the shipped wording, assigned by a stable hash of the source image path and recorded as metadata.query_template (37 templates in use, top share 3.4%).

Template 0 is v1's wording byte for byte (124 records keep it); the pass asserted that on every record before rewriting anything.

Template ↔ gold independence on this build: 4,798 records, 37 templates, worst template p = 0.00453, alpha 2.7e-04, 0 flagged → PASS.

Frame-size floor (common/lazy_floors.py, the standing (width, height)-only row): vacuous by construction — all 4,798 images share one frame size.

Answers, images, masks, split and every other field are byte-identical to v1: this revision was issued from the published parquet itself (tools/requery_published.py), not rebuilt from source, and the pixel-identity guard ran on the embedded images (§8 below).

Provenance

Underlying dataset: CableInspect-AD. Upstream license: CC BY-NC-SA 4.0 (Hydro-Quebec; NeurIPS 2024 D&B) (this card is license: other; respect the upstream terms). Converted read-only from the raw source into the unified schema; conversion code under 208/ (with publish/push_to_hf.py) in AI4Manufacturing/forge_model.

Overlap / de-duplication (§8)

No overlap with any other dataset in this corpus — this is field maintenance inspection, not production-line imagery. ⚠ There is no official split and folds must be cable-wise — see the split note below.

Two identities, and they answer different questions. metadata.image_sha256 hashes the file bytes: it finds byte-identical copies and is blind to a re-encode. metadata.pixel_sha256 hashes the decoded image (mode | size | pixels): it finds the same photograph saved twice. Only the second one settles whether an image is duplicated.

Measured at build time, not asserted afterwards — a violation aborts the build and names the offending records:


images checked	4,798
distinct by decoded pixels	4,798
images carrying more than one record	0
images on both sides of the split	0

Geometry (`metadata.geometry`)

Every record carries a geometry block inside the existing metadata JSON string, so that its gold can be re-derived at any render size. No schema column changed; existing loaders are unaffected.

Coordinates are native pixels of the image in that record (coords_frame: "record_image"). scale is 1.0 throughout — this repo publishes at source resolution, nothing was downscaled at publish time.

"geometry": {
  "image_wh":  [W, H],        // dims of the image in THIS record
  "source_wh": [W, H],        // dims of the original source image
  "scale": 1.0,               // image_wh / source_wh; < 1.0 would disclose a publish-time downscale
  "n_instances": 2,
  "instances": [
    { "instance_id": 1, "bbox_xywh": [x, y, w, h], "min_side_px": 65, "class": null }
  ],
  "n_dropped_subminimum": 0,  // components removed by the filters below
  "union_box_fallback": false,// true => boxes are per-class unions, NOT real instances
  "conventions": { ... }      // see table
}

instances is present even when empty. [] means the record genuinely has no defects; an absent block would mean geometry could not be recovered. Those are different states and are never conflated.

Conventions used to derive it

There is no universal definition of "one defect instance" — it depends on the mask the source shipped. This repo's is stated, not implied:

field	value
`algorithm`	`source_annotation`
`binarisation`	`n/a`
`connectivity`	`4`
`merge`	`none`
`min_area_px`	`0`
`max_instances`	`None`
`artifact`	`fine`
`fill_floor`	`None`
`legibility_floor_px`	`None`
`min_side_floor_px`	`None`
`spec_sha`	`79a4dea70a32ebf9`

Provenance and verification


records	4,798
carrying a geometry block	4,798 / 4,798
instances per record	`0`: 2,159, `1`: 1,783, `2`: 554, `3`: 243, `4`: 51, `5+`: 8
total instances	3,864
image dimensions	1920×1080 (4,798)
`scale` values present	[1.0]

Computed from this repo's own masks and verified against this repo's own published answers before it was written — a recomputation that disagreed with the shipped gold would have aborted the update rather than overwritten it.

⚠ The 16px floor applies at the RENDER, not at native

min_side_px is in native pixels. The model does not see native: Qwen2-VL caps by megapixels AND snaps each dimension to a multiple of 28. So min_side_px >= 16 is the floor tested in the wrong frame. Measured on this repo:


native → rendered (qwen2_vl @ 2.36MP)	1920×1080 → 1932×1092
shipped boxes	3,864
legible at that render (>=16px there)	3,862 (99.9%)

⚠ An earlier version of this section reported the inverse — boxes clearing 16px at native and failing at the render — and that number was misleading. It is frame-relative: publishing at a larger native size lets more boxes clear 16 in the published frame, so more can "fail", which penalises exactly the choice that helps. Measured on 179: publishing native (3024) means a box needs >=32px native to be legible at the render and 86.7% qualify; the previous 1024 publish needed >=47px native and only 69.5% qualified. The native republish improved rendered legibility by 17 points while the old metric scored it as 12.5% "broken". The figure above is the comparable one.

Nothing in the data is frame-dependent — geometry is native and complete. Use forge_model/common/adapt_engine.py, which applies the floor at whatever size the consumer renders.

Using it

Coordinates only stay correct if they are rescaled with the image. A patch-based VLM does not render at native size: Qwen2-VL's processor snaps both dimensions to a multiple of 28, so this repo's 1920×1080 is rendered 1932×1092 and native-pixel boxes are then wrong by a few pixels. forge_model/common/adapt_engine.py regenerates coordinates for a target render size, re-derives counts, and drops records whose gold no longer holds there.

Downloads last month: 99

Total file size:

13 GB

You need to agree to share your contact information to access this dataset