{Section Name}

One-sentence description of what this CV subtask does and what changed recently in the field.

Two paragraphs of context: - What this subtask is and where it fits in the pipeline. - Why the picks differ (what axis separates them — hardware, accuracy tier, deployment target).

Recommended picks

Tier	Pick	When to use
Edge / low-end	Name	One sentence on the constraint.
Default	Name	One sentence on why this is the first try.
Max accuracy	Name	One sentence on the unconstrained-compute scenario.

Why {default pick} is the default

1-2 paragraphs. Include: - Published year / maintainer / origin. - What makes it the default in 2026. - pip install / repo link / how to try it in 5 minutes.

Why

Same structure.

Why

Same structure.

When to pick something else

Bullet list of specific scenarios where none of the 3 picks above are right, and what you'd reach for instead. Each bullet should have a why.

• Scenario → Alternative pick and why.
• Scenario → Alternative pick and why.
• Scenario → Alternative pick and why.

The three questions to narrow

1. Question 1? Answer → Pick X. Answer → Pick Y.
2. Question 2? Answer → Pick X. Answer → Pick Y.
3. Question 3? Answer → Pick X. Answer → Pick Y.

The Dump

Everything else worth knowing in this subtask. One-line verdict per entry. Group by category if helpful:

Category A

• Name (year) — one-line verdict.
• Name (year) — one-line verdict.

Category B

• Name — one-line verdict.

Tools / libraries

• Name — one-line verdict.

Graveyard

• Name (year retired) — why it was the pick; why it's retired.
• Name (year retired) — why it was the pick; why it's retired.

Last reviewed

YYYY-MM-DD.