Screenshot classification
Before
browser_dashboard.png with a browser-dashboard visual caption
After
Screenshots category
Direct answer
On September 6, 2026, Galoria passed all 26 backend tests in 1.986 seconds on Windows 11 Pro with a Core i5-8350U and 8 GB of RAM. A representative regression set correctly handled all 12 defined category cases, including work, people, travel, products, receipts, documents, screenshots, wallpapers, design assets, memes, and pets. This is a curated regression result, not a claim of 100% real-world accuracy.
Reviewed and updated 2026-09-06
Observed Galoria test evidence
The regression results and product captures document both machine-checked behavior and the user-visible workflow.
- Backend tests
- 26 passed, 0 failed
- Representative fixtures
- 12 passed, 0 failed
- Automated suite time
- 1.986 seconds
- Environment
- Windows 11 Pro, Core i5-8350U, 8 GB RAM
No general accuracy percentage or guaranteed scan speed is inferred from these fixtures. A broader benchmark would require a labeled image dataset, repeated runs, and documented model/runtime versions.
Why photo-organizer evidence needs edge cases
A useful image organizer must do more than match obvious words. Galoria's regression suite checks that a camera photo of a window is not treated as a screenshot, an abstract camera photo is not forced into Wallpapers, and a small unknown image can remain uncertain instead of receiving a confident but weak category.
The suite also covers preview capabilities, changed-file rejection, duplicate verification with a full hash, undo history, bounded progress updates, metadata workers, and caption limits. These checks reduce known regressions while leaving the final destination visible to the user.
Step-by-step workflow
- Prepare representative filenames, image metadata, and optional local visual captions with an expected category for each case.
- Run the classifier across 12 categories covering common photo-library and screenshot workflows.
- Run edge cases designed to catch screenshot, wallpaper, design, and uncertainty false positives.
- Run planner, duplicate, history, preview, changed-file, and performance-bound tests before publishing the result.
Use-case examples
Receipt classification
Before
receipt_204.jpg with printed receipt, subtotal, and tax signals
After
Receipts category
Evidence, meaning, and limits
The published result separates curated regression coverage from performance and real-library accuracy.
| Option | Best for | Tradeoff |
|---|---|---|
| 12 category fixtures | Preventing regressions in representative filename, metadata, and caption cases | A curated set is smaller and cleaner than a real photo library |
| 26 backend tests | Classifier, planner, duplicate, history, preview, and bounded-work checks | Test runtime is not end-to-end library speed |
| Copied-library acceptance test | Checking categories against your own image collection | Requires reviewing ambiguous and irreplaceable images |
Frequently asked questions
Did Galoria score 100% accuracy?
Galoria passed 12 of 12 defined regression fixtures. That is not the same as 100% accuracy across arbitrary real-world photo collections.
What happens when an image is uncertain?
The classifier can retain uncertainty rather than forcing a weak category, and the proposed destination remains available for review.
Does the 1.986-second result measure a photo scan?
No. It is the complete backend automated test-suite time on one computer. A real scan also depends on image count, format, storage, caching, and visual analysis.
Related guides
Organize a folder with Galoria
Run local AI on Windows, review the proposed structure and filenames, then apply the changes you approve.
Download for Windows Get the complete bundle