INDUSTRY FIELD GUIDE
Cookie and cracker data for AI breakage and occlusion review
A cookie or cracker line can contribute visual examples whose labels state exactly what the camera shows.
NAICS 311821 · Cookie and Cracker Manufacturing
A cookie or cracker line can contribute visual examples whose labels state exactly what the camera shows. NIST's packaged-goods handbook provides measurement context, while the proposed exercise concerns visible breakage and incomplete views. NIST Handbook 133
Arrange original inert shapes in a tray and photograph a complete view, a flap-obscured view and a separate set of broken pieces. Provide a fictional packing sheet that identifies the intended count. Ask the assistant to report visible whole pieces and the evidence needed to reconcile the pack. Do not let the answer key equate the number of fragments with the number of original pieces.
Annotate each visible item, occluded region and uncertain association. A second camera angle should resolve some cases while leaving others incomplete. Hold out a new tray pattern for evaluation and score invented hidden pieces separately from ordinary counting mistakes. This creates a task in which requesting a better view is a valid outcome.
Begin with props and original packaging. Real customer molds, distinctive designs and line footage need authorization before licensing. Describe whether a prospective buyer wants image segmentation, pack reconciliation or both. The capture angles and review key should follow that accepted task, so additional examples answer a defined need.
Discuss this original manufacturing sample with Fiund
Original Fiund exercise
Several visible fragments do not identify the number of original pieces
An original tray of inert cracker-like shapes has reviewed textual view annotations. It contains no food, actual imagery, production line or package-quantity claim.
| Record | Source | What it records |
|---|---|---|
| CBV1 | Packing target | Practice carton K intended to contain six whole shapes. No verified original pack count stated. |
| CBV2 | Complete front view | Four whole shapes and three disconnected fragments visible. Reviewer cannot link fragments to their original parents from this angle. |
| CBV3 | Flap-obscured view | A second angle shows three of the four whole shapes and hides the right compartment. No added evidence about fragment identity or unseen contents. |
Question for the model
What visible count can be reported, and can the fragments or the obscured view establish six original pieces?
Show the reference answer and checks
CBV2 supports four visible whole shapes and three fragments, not seven whole pieces or a known original count. The fragments lack a verified parent grouping. CBV3 shows fewer visible shapes because of occlusion and does not demonstrate removal or unseen contents. CBV1 is an intended count, not ground truth. Report separate visible categories and request a view or reviewed fragment association rather than force a reconciliation to six.
- Counting each disconnected fragment as an original piece.
- Interpreting fewer visible objects in an obscured view as a true inventory change.
Fields to preserve
- Carton and intended count
- Defines the target.
- Whole-object and fragment annotations
- Supports separate counts.
- Occluded region and fragment parent
- Limits reconstruction.
Test another case
Use a new layout where two fragments have a reviewer-confirmed shared parent. Require that pair to count as one linked parent, with other fragments unresolved.
Use original inert shapes and packaging. Exclude customer molds and actual production media without 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.