Datasets:
You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This dataset is released for research use. Access is reviewed and granted manually by the maintainers. Please state your name, affiliation, and intended use.
Log in or Sign Up to review the conditions and access this dataset content.
Roles
Roles: canon repo — annot is the source label (good / anomalous, <type> at OBJECT level), kept machine-parseable as the gold for verification and reward parsing; images holds the five views the model reads and image is a preview of the first one; masks are per-view GT and are not model input; the reasoning column is empty, so this repo is not itself a training view for chain-of-thought. metadata is provenance (split keys, geometry, identities) and must never be fed to a model.
192-object
Multi-view industrial anomaly detection — OBJECT (sample-level) version (5 views per row; object-level binary; per-view masks kept as 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
30,210 records (test=22917 · train=7293).
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: the plain-text object-level label good or anomalous. One row = one object = five synchronised views in images (C1 top + C2-C5 side); the object is anomalous iff it is an NG sample even if some views look good (invisible-view effect; per-view labels in metadata.views). The per-view masks are aligned in masks (None where a view has no defect) but NOT referenced in the query — see Versions, 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] | (multi-image T-B1 / D21) per-view masks aligned with images (None where a view has no defect), or multi-region masks |
Versions, task & split
What this is. Real-IAD (Wang et al., Real-IAD: A Real-World Multi-View Dataset for Benchmarking
Versatile Industrial Anomaly Detection, CVPR 2024) — 30 manufactured objects, five synchronised camera
views per object (C1 top-down + C2-C5 at 45°), with pixel defect masks. 256-px release
(realiad_256).
Two published versions (same images, different unit).
- 192-object (this repo) — one object (its 5 views) per row, sample-level binary AD (Real-IAD's headline S-AUROC): the five views are aggregated into one object decision.
- 192-single — one camera view per row, image-level binary AD (I-AUROC). They are the same photos re-grouped — keep the two on the same side of any train/eval split.
One row = one object. images holds the object's five views in order C1..C5 (image scalar = the
C1 top-down view, a preview); masks holds the per-view defect masks aligned with images (None where
that view has no defect), attached but not referenced in the query.
Query & answer. query (our own template, per category) asks whether the object is good or
anomalous; annot is the plain-text label. The object is anomalous iff it is an NG sample — even if
some of its views individually look good (the invisible-view effect: a defect may be visible from only
one or two angles). The per-view labels (which view is good/anomalous, whether it has a mask) are in
metadata.views; the object-level defect type is metadata.defect_code / defect_name (Real-IAD's 8
types AK/BX/CH/HS/PS/QS/YW/ZW).
Split. The main split is published: train = normal-only objects (unsupervised-AD protocol),
test = mixed. ~30,210 objects total (151,050 images / 5 views). (Real-IAD also defines FUIAD
noisy-train splits; only the main split is published.)
Query text — pooled paraphrases (v2)
Every record's query is drawn from common/vision_query_pools.json[F3/verdict_word], a pool of 40 gate-verified paraphrases of the shipped wording (this repository draws from the 10-template family of the ask it shipped; the other families describe inputs of a different shape), assigned by a stable hash of the source image path and recorded as metadata.query_template (10 templates in use, top share 10.2%).
The opening role sentence is drawn separately (metadata.query_role, a 10-way hand-written pool _role/sentence; index 0 is this repository's own sentence, index 1 is none); the subject sentence is this repository's own, verbatim, on every record. Role and ask are hashed independently.
Template 2 is v1's wording byte for byte (310 records keep it); the pass asserted that on every record before rewriting anything.
Template ↔ gold independence on this build: 30,210 records, 10 templates, worst template p = 0.197, alpha 1.0e-03, 0 flagged; 10 roles, worst role p = 0.00886, 0 flagged → PASS.
Frame-size floor (common/lazy_floors.py, the standing (width, height)-only row): vacuous by construction — all 30,210 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: Real-IAD. 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 code under 192/ (with publish/push_to_hf.py) in AI4Manufacturing/forge_model.
Overlap / de-duplication (§8)
Same underlying images as 192-single (this is the multi-view grouping) — keep the two on the same side of any split. Published main split only.
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 | 181,260 |
| distinct by decoded pixels | 151,049 |
| images carrying more than one record | 1 |
| images on both sides of the split | 0 |
images checked counts every entry of images plus the image preview column, which is a copy of the first view — 30,210 records, 181,260 decoded images; the distinct count is therefore lower by one per record by construction, and only the more than one record row measures duplication.
⚠ This dataset declares a many image-identity policy, so the row above is expected to be non-zero: UPSTREAM (Real-IAD 256 release), the same defect 192-single declares: one file is shipped twice under two different objects with contradictory labels — view C1 of button_battery/OK/S0183 (object labelled good) and view C5 of phone_battery/NG/HS/S0039 (object labelled anomalous) are byte-identical (file hash f256db31…), both in test. Both object records are kept exactly as published; one of the two labels is wrong. Recorded for the next data revision Images are still forbidden from crossing the split, and none do.
- Downloads last month
- 53