Where in the world?
Geo-Estimation
Establish a candidate region from the photograph’s visual content, without relying on metadata.
AvailableOceanir / Manta
A location is
a claim.
Show the evidence.
Manta turns visual clues into locations you can examine.
Move from a candidate region to a street, building, or interior match. Compare the source imagery, inspect the evidence, and keep uncertainty in view.
For approved organizations and reviewed workflows.
Geo-EstimationSource photograph → nearby reference imagery
Inspect the leading estimate against nearby imagery. This example remains unconfirmed; the reference view gives you context to review the location.
CapabilitiesFrom a place
From a place
to the proof.
Three modes for different stages of an investigation. The precision of the answer is limited by the evidence behind it.
Which street or building?
Ground
Compare the scene with reference imagery to resolve a candidate location at street or building level, where coverage supports it.
AvailableDoes the evidence match?
Verify
Inspect interior listing candidates in a selected city or region, compare photographs, and examine the visual correspondences.
Reviewed previewVerify / Workspace
Inspect an interior candidate.
Compare listing photographs and matching points inside Workspace. This saved Miami example remains Not verified: a point count alone does not confirm a property.
Read about Verify
EvaluationMeasure the model.
Measure the model.
Then test your case.
Public benchmarks describe performance on their test sets. They do not establish the accuracy of a particular photograph or a Verify result.
Ground / Nordland
97.21%
Recall@1 · +1.21 points vs SAGE (8448-D)Across seven VPR benchmarks
2 / 7
Numerical leads · #2 of 11 by mean Recall@1Geo-Estimation / OSV-5M
30.5 km
Median error · 32 km on the Chipoint-2 cardInspect all benchmark results
| Benchmark | Ground | Best other reported | Margin (pp) |
|---|---|---|---|
| SPED | 95.39 | SAGE (8448-D) · 98.9 | −3.51 |
| Pitts30k | 94.95 | SAGE (8448-D) · 95.8 | −0.85 |
| MSLS val | 94.60 | SAGE (8448-D) · 94.5 | +0.10 |
| Nordland | 97.21 | SAGE (8448-D) · 96.0 | +1.21 |
| AmsterTime | 74.25 | SAGE (8448-D) · 83.5 | −9.25 |
| Tokyo 24/7 | 98.43 | EffoVPR · 98.7 | −0.27 |
| SVOX | 98.74 | BoQ · 99.0 | −0.26 |
Geo-Estimation / OSV-5M / share within distance
45.77%
Within 25 km
+0.94 pp vs Chipoint-281.83%
Within 200 km
+0.73 pp vs Chipoint-292.57%
Within 750 km
+0.46 pp vs Chipoint-297.30%
Within 2,500 km
+0.48 pp vs Chipoint-2Numerical leads, not established improvements beyond evaluation variability. Sources, comparisons and protocol are in the full evaluation.
Read the evaluationAccess & limitsBuilt for review.
Built for review.
Scoped to the work.
Manta access is granted to an organization, with a declared purpose, named users, and a person accountable for the conclusion.
Start with an organization evaluation.
Enterprise organizations, selected pilots, and research partners can request access. Oceanir’s standard model remains the default in the app and API.
- Reference geography
- Worldwide
- Ground precision
- Street and building matches worldwide
- Verify preview
- Available in any selected city or region
- Access
- Scoped by organization and declared purpose
Some images stay unresolved.
- Indistinct interiors
Rooms with few distinctive fixed features can leave Verify unresolved.
- Missing visual clues
Overcast skies remove the light it reads latitude from.
- Similar landscapes
Motorways across temperate countries can look alike. Sometimes the evidence supports only a region.
The people boundary
Manta establishes where a photograph was taken. It is not built to find people, and we do not permit it to be used that way.
Evaluate MantaBring a case.
Bring a case.
Examine the answer.
Include photographs with known locations. Review what Manta produces, including where it declines to commit.