Industries
Not because the model is weak, but because it has never seen what it now has to handle. These six sectors are where the gap between foreign data and Vietnamese reality is widest.
Traditional trade drives most retail volume in Vietnam, yet almost no imagery of it exists. Every off-the-shelf shelf dataset was photographed in a Western supermarket.
Motor claims are the largest data asset the insurance industry is not using, because scene photography has no labels. It is also the sector where we already have a model in production.
Vietnamese roads are dominated by motorbikes and use signage that differs from international standards. A model trained elsewhere cannot handle that.
Vietnamese documents carry diacritics, handwriting, faded print and skewed captures — everything that breaks extraction models trained on printed English.
Pests, cultivars and growing seasons in Vietnam differ entirely from the common agricultural datasets, which were collected in temperate regions.
Research groups usually need very specific data without a corporate budget. We hold a separate academic license for them.
These six are where we work most, not a limit. If your problem needs data captured in Vietnam, describe it — we will say plainly if we cannot do it.