Retail and warehouse perception systems work against a catalog that
never stops changing: a segmenter deployed today will eventually be
asked to find products that did not exist when it was trained. A closed-
set segmenter can only grow its vocabulary by widening its classification
head and fine-tuning it, which risks the classes it already handled, or
by retraining on the whole accumulated catalog, which costs more with
every product added. This thesis proposes reference-prompted segmen-
tation, which specifies the target product with a reference prompt, a
fronto-parallel view of the product, rather than a class label or a text de-
scription, so that a product’s identity enters the model as an input rather
than as a slot in its output. We fine-tune a reference-prompted segmenter
one new product at a time — twenty-seven sequential steps growing a
three-product catalog to thirty — using Learning without Forgetting,
and compare it against a closed-set segmenter given the identical recipe,
budget, and data. Our method retains its earlier products substantially
better than the closed-set baseline. Every dataset in this thesis comes
from isaac_datagen, a standalone Isaac Sim synthetic-data engine of-
fered as a reusable contribution in its own right. Under this protocol and
at this scale, the reference-prompted segmenter retains 5× the panoptic
quality of an equally-trained closed-set segmenter after sequentially ab-
sorbing 30 products (0.275 vs. 0.054 PQ), letting a perception system
keep pace with a growing catalog without ever retraining on all of it.