VISIONROUTE

Guide · Pillar: Prove

It depends — here are the variables and honest ranges.

For evaluation, you need fewer images than teams fear: 100–300 well-chosen examples expose most decisive differences between approaches. For training, ranges span orders of magnitude: zero for a prompted VLM, tens per class to start moving a fine-tuned model, hundreds or thousands per class for subtle distinctions. Any single number quoted without your task’s variability, class count, and accuracy bar attached is a guess — the table below states its uncertainty instead of hiding it.

Published August 31, 2026 · Pillar: Prove · By the VisionRoute team

Definitions

Two different questions

"How many images do I need?" conflates two budgets. The evaluation set proves whether any approach meets your bar; the training set teaches a model your task. Every project needs the first; only some architectures need the second — which is why the table has two count columns.

Representative

An image set is representative when it matches production reality: the same capture conditions, the same mix of easy and hard cases, and true negatives. Two hundred representative images beat two thousand scraped lookalikes — count is the weakest property of a dataset.

Who this is for

Teams budgeting data collection or labeling before committing to an approach — and anyone quoted a single confident number by a vendor or a forum thread and wondering why the range is so wide. The scope is still-image product decisions.

Why ranges, not numbers

Published rules of thumb for image counts disagree by orders of magnitude, because task difficulty, class similarity, and variability dominate the arithmetic. That disagreement is the finding: treat every count — including ours — as a starting budget attached to a stopping rule.

The scenario table

Ranges by scenario, with the uncertainty stated

These are starting budgets drawn from practitioner experience, not measurements — the fourth column says how much to trust each row. Terms are in the glossary.

Starting budgets for still-image product decisions. The stopping rule below — not this table — decides when you actually have enough.
ScenarioTraining imagesEvaluation imagesConfidence in this range
Prompted VLM, no training0 — plus 5–10 exemplars if you use few-shot prompting.100–300, built backwards from failure.High that zero works to start — that is the point of prompting. The evaluation need never goes away.
Specialist API, used as-is0.100–300, weighted toward your hard cases and negatives.High — but "the API covers your task" is exactly the claim the evaluation set exists to check.
Fine-tune a pretrained classifier, visually distinct classesRoughly 30–100 per class before results move.50+ per class you care about.Medium — class similarity and image variability swing this by multiples in both directions.
Fine-tune for subtle distinctions (defects, grades, lookalikes)Hundreds per class, often more.100+ per decisive class, borderline cases included.Low–medium — subtlety is precisely what inflates data needs. Expect iteration, not a number.
Fine-tune a detector on custom classesTens to hundreds of labeled boxes per class.50+ images per class, including crowding and occlusion.Medium — label quality (tight, consistent boxes) matters about as much as label count.
Train from scratchMany thousands per class.A large held-out set on top.High that you should not start here — without a specialist team and evidence that pretrained models fail your task, scratch training is a research project.

The stopping rule

Collect until the decisive metric stops moving

  1. Label the evaluation set first — always

    The first 100–300 labeled images become the evaluation set, untouched by training. Teams that train on everything and evaluate on leftovers end up grading the model on its own homework.
  2. Grow training data in increments

    Start near the low end of your scenario’s range, then roughly double: 50, 100, 200, 400. Retrain and re-evaluate at each step, watching the decisive slice — not the average.
  3. Stop when the curve flattens

    When the decisive-slice metric stops improving across two consecutive increments, more of the same data has stopped paying. Spend the remaining labeling budget on rare slices, hard negatives, and fresh failure cases instead.
  4. Let the costly class set the budget

    With imbalanced classes, the average is a decoy. If the rare class is the expensive one — the defect, the violation — its count is your binding constraint, and "enough images overall" can still mean "not nearly enough of the ones that matter".

When this changes

More classes, more budget

The per-class counts above multiply by your class count, and confusable class pairs need extra examples along their shared boundary. Twenty subtle classes is a different project than three distinct ones.

A higher accuracy bar

Each step toward "almost never wrong" costs disproportionately more data and iteration than the last. If the bar is extreme, revisit whether an abstention path can carry part of it before data collection does.

High variability capture

User-submitted photos from arbitrary phones need budgets at the top of every range; a fixed camera on a fixed line sits at the bottom. Variability, not task type, is often the biggest multiplier.

Synthetic augmentation

Augmented and generated images can stretch a small set, especially for rare cases — but they encode your assumptions, not reality. Whatever they contribute, the evaluation set stays real images only.

Limitations

What this table cannot promise

  • No universal number exists, here or anywhere. The ranges order scenarios relative to each other; your task can still land outside its row, which is what the stopping rule is for.
  • The table assumes competent labels. Label noise quietly raises every count — disagreeing annotators are a definition problem to fix before a volume problem.
  • It budgets data, not outcomes. Whether the resulting stack meets your bar is settled by evaluation — see how to test on your own data.

In short

Evaluation needs 100–300 well-chosen images for most narrow tasks; training needs anywhere from zero (prompted VLMs) to tens or hundreds per class (fine-tuning) to thousands (subtle distinctions or scratch training). Treat every count as a starting budget with a stopping rule — grow data in increments and stop when the decisive-slice metric flattens — and let the rare-but-costly class, not the average, set the budget.

Related: the feasibility checklist · test models on your data · feasibility testing explained