Short answer
Product catalogs change constantly: new SKUs launch, packaging is redesigned, items are retired. A one-time image purchase goes stale, so recurring supply needs its own contract shape: a defined refresh scope, versioned deliveries, stable identifiers, rules for retired items and clear terms on whether earlier versions may stay in trained models. This is not legal advice.
Why a refresh needs different terms from a snapshot
A product-recognition model trained on last year's catalog meets this year's shelf with a share of items it has never seen, and a further share that look different because the same product ships in new packaging. For a humanoid working in retail, warehousing or kitting, that drift is a standing maintenance cost, not a one-off event.
A snapshot license answers a single question: what may I do with this file set? A recurring supply has to answer more. What will be delivered, how often, in what format, tied to which identifiers? What happens to items that leave the catalog? Do earlier deliveries stay licensed if the relationship ends? The contract needs to reflect those moving parts, and the supplier's internal process has to be able to meet them.
This page covers the licensing and delivery side. For what catalog data contains and what it can support, see product catalog data for robot perception, and for how packaging redesigns alter appearance, packaging variants and SKU appearance. The cluster overview is the product and object data hub.
What a catalog snapshot looks like
ABO illustrates the structure of a catalog-derived dataset: it pairs product listings and metadata with images and 3D models, with 147,702 product listings, 398,212 unique catalog images, up to 18 metadata attributes per product and 7,953 products with artist-designed 3D meshes (Collins et al., 2021). It also provides 360-degree turntable views for 8,222 products.
That is a snapshot of one catalog at one time, built for research. It says nothing about how often a catalog changes or how a supplier would deliver updates, so any refresh cadence below is a design choice for your contract, not a figure drawn from the paper.
Clause topics for recurring supply
Scope of each delivery
State whether a refresh covers new SKUs only, changed SKUs, retired SKUs, or the full catalog each time. Define "changed": a new front panel, a new barcode, a new net weight, a recolored cap. Agree which attributes are committed to each refresh and which are best-effort. See the attribute schema page for fields worth fixing in advance.
Identifiers and versioning
Without stable identifiers, a refresh cannot be reconciled with what you already hold. Require a persistent product ID that survives repackaging, a version or valid-from date on each image and attribute row, and a mapping when a SKU is merged or split. Ask for a changelog with every delivery.
Retired and recalled items
Decide whether retired items stay in the delivery, flagged as inactive, or are dropped. Retired products still appear on shelves and in returns streams for a while. A recalled item raises a separate question: whether the supplier will notify you and whether you must retrain or remove it.
Rights through time
Image rights, brand rights and photographer rights may differ per refresh batch. Ask whether each delivery is licensed on the same terms or renegotiated, and whether earlier versions in trained models can be kept if the supply stops. Also ask what the supplier does when a photographer or brand withdraws permission for an item. See product image rights for the underlying permission chain, and the broader licensing terms for robotics training data.
Permitted uses and restrictions
Specify training, evaluation and internal derivative use, and confirm whether outputs, such as embeddings or fine-tuned weights, are covered. Restrictions on competing uses or on redistribution belong in the license, not in an email. SourceX's own summary of terms is on the licensing terms page, and a general explainer is in how to license proprietary data for AI training. This is not legal advice; have counsel review the draft.
Illustrative delivery schedule
Illustrative only. The cadence, fields and thresholds below are arbitrary examples of what a buyer might negotiate, not recommendations, norms or minimums.
| Element | Illustrative term | What to confirm |
|---|---|---|
| Delivery interval | quarterly | Whether the supplier's own update cycle supports it |
| Content per delivery | new and changed SKUs plus full attribute table | Definition of "changed" |
| Identifier | product_uid persistent across versions | Behavior on merge or split |
| Version tag | valid_from date on each row | Whether history is retained |
| Retired items | flagged inactive, kept for one cycle | Whether returns and secondhand flows matter to you |
| Evaluation set | frozen, never refreshed into training | Who controls the freeze |
Delivery mechanics
Settle format and transfer method up front: image files, attribute tables, 3D assets and a changelog, delivered to a defined location with checksums. Decide who validates each delivery and how long you have to report problems. If part of the supply comes from several suppliers, make sure identifiers line up across them, which the product master data quality audit page addresses.
Keep your evaluation set frozen. Refreshes should feed training only, with the evaluation set held out by item, brand or time as your claim requires. Mixing new items back into the test set quietly changes what you are measuring.
What a refresh contract cannot guarantee
A refresh agreement cannot make a supplier's catalog change more regularly than it does, and it cannot supply products the supplier never lists. It does not capture how items look when worn, crushed or half-hidden on a shelf: catalog images are not in-situ images, and a refresh brings more of the same kind. It does not add handling information such as weight distribution, rigidity or slip. Those need robot or teleoperation data. See packshot versus in-situ images.
Also be careful about three things:
- Continuity. If the supplier is acquired, changes platform or stops shooting new products, the supply can end. Ask for a notice period and a clear end-of-supply clause.
- Coverage drift. New items may skew toward categories the supplier is expanding in, not the ones your site stocks.
- Quality drift. Image conventions change over time: backgrounds, angles, resolution. Ask for a style guide and change notes.
Whether the supplier is a retailer or the maker also changes what a refresh can contain; see retailer versus manufacturer object data.
Questions to ask suppliers
- How often does the catalog change, and how are changes recorded internally?
- Is there a persistent product ID, and what happens to it when a SKU is merged, split or repackaged?
- Can each image and attribute row carry a valid-from date?
- Will retired and recalled items be flagged, and will you notify us?
- Are rights to new images granted automatically under the same license, or per batch?
- May earlier versions remain in trained models if the supply ends?
- Who validates a delivery, and what is the process for correcting errors?
- Do image conventions (angles, background, resolution) change over time, and how will that be documented?
- Who at the supplier can authorize the license for photographs, 3D models and brand marks?
Discuss your requirements
If you need product imagery that stays current, describe the catalog scope, refresh pattern and permitted uses to SourceX. SourceX looks for US businesses that hold suitable material, checks the data and the supplier's licensing permissions, agrees pricing and allowed uses in a license, and manages future purchases. A request does not guarantee a match; SourceX follows up about potential sources. The humanoid training data guide shows the other sourcing routes, and the catalog entry for product catalogs and descriptions describes the related data type.
Explore sourcing and licensing options with SourceX
Sources
- ABO: Dataset and Benchmarks for Real-World 3D Object Understanding — structure and size of one catalog-derived snapshot; not evidence about refresh practice.