Coarse-to-Fine
lastmod 2026-03-10created 2026-03-094 edits
The coarse-to-fine strategy is a hierarchical approach to solving complex problems that
- first grasps the rough (Coarse), low-resolution overall structure,
- then gradually incorporates fine (Fine), high-resolution details
to increase precision. It is mainly used in AI image processing, 3D modeling, brain mapping, and similar areas, achieving both computational efficiency and accuracy.
Pronunciation
Written in Hangul, the pronunciation of 'Coarse' is close to "코스" (koseu) or "코얼스" (koeolseu).
- American English: '코얼스' (koeolseu) (the r sound comes in slightly, with the tongue rolled)
- British English: '코스' (koseu) (drawn out long, with no r sound)
For reference, it is pronounced exactly the same as the familiar word 'Course' (a process, a path).
Key ideas and application examples
Solves problems by gradually narrowing down from large chunks (Coarse) to small parts (Fine).
- AI and computer vision
- Processes images from low to high resolution to reduce computational cost -> deblurring (Deblurring)
- Classifies objects from low to high resolution, as in Coarse-to-Fine Grained Classification.
- Approaches like HierDiff first establish the large structure of a molecule and then assemble the detailed atoms; in brain mapping, mapping proceeds from large brain regions down to detailed subregions.
Pros
Lets you efficiently refine (Refinement) down to the details (Fine) without losing sight of the overall structure (Coarse).