INDUSTRY FIELD GUIDE
Textile roll inspection data for woven and knit mills
An isolated image of fabric rarely tells an inspector where to find a fault on the roll.
NAICS 313210 · Broadwoven Fabric Mills
NAICS 313240 · Knit Fabric Mills
An isolated image of fabric rarely tells an inspector where to find a fault on the roll. A useful dataset associates the image with the roll position, edge distance, and fabric construction. NIST’s NIR-SORT project publishes a distinct kind of textile data, near-infrared spectra for fiber classification. It illustrates why labels and material variation matter. A visual fabric-defect project would need its own labeled images and task scope. NIST NIR-SORT · NIST manufacturing defect segmentation
Try a short training roll with a documented snag and an intentional texture feature. Record distance along the roll and across its width. Include an overview with a scale and a close-up of the feature. For the woven set, add an example across the warp or weft. For the knit set, include how the same area looks under a consistent relaxed condition. An evaluator can ask whether the model found the location, whether it knew that a visual irregularity was intentional, and whether the location could be reproduced at inspection.
Use owned patterns or neutral material. Share measurements under anonymous roll IDs, and keep customer designs, private fiber formulations, and mill process settings out of the sample.
Describe your roll inspection data
Original Fiund exercise
A woven repeat and a local strand break
A contributor constructs an original woven swatch with a repeated raised motif and a simulated discontinuity. The model labels visible evidence only.
| Record | Source | What it records |
|---|---|---|
| T1 | Reference strip | Raised horizontal bars recur every 40 mm; bar edges remain continuous. |
| T2 | Roll photo index | At 80 mm the raised bar is continuous and matches the reference. |
| T3 | Detail pair | At 95 mm a lifted strand interrupts an otherwise straight run; mark visible in both views. |
| T4 | Capture note | At 120 mm one view is obscured by glare and there is no second view. |
Question for the model
Which locations should be labeled as expected texture, a visible discontinuity or unresolved?
Show the reference answer and checks
Label 80 mm expected texture using T1 and T2, and 95 mm a visible local discontinuity using T3. Leave 120 mm unresolved from T4 because glare removes the relevant evidence. This identifies what is visible without assigning root cause or a roll grade.
- Learning that all raised features are defects
- Treating glare or missing imagery as an inspected clean segment
Fields to preserve
- roll_offset_mm
- Places an observation on the long axis
- reference_pattern
- Distinguishes repeated geometry from a local departure
- view_evidence
- Records whether the relevant region is visible in each view
Test another case
Rotate a striped swatch and place a simulated break inside a motif; group the holdout by swatch so near-duplicate crops cannot leak.
Use an owned swatch and staged imagery; no real buyer grading sheets or third-party fabric design rights.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.