- Take photos
- Open CAD or measurement software
- Manually trace and measure
- Prepare report evidence
Falcon Eyes Research + Product Company
Measure real-world objects from a single image.
Take a photo of a crack, pipe, room, machine part, or field asset. GaugeAnything turns visual perception into measurement outputs: millimeters, counts, spacing, confidence, and inspection-ready evidence.
Demo first
Upload Image
Secure endpoint pending
Target
Building
Output
Crack width, bolt count, spacing
Workflow time
Seconds
Claim boundary
Measured with scale and uncertainty checks
Problem to solution
The new inspection workflow is measured, not manually annotated.
If an AI pipeline ends at a mask, the field team still has to do the expensive part: turning that mask into a number that can drive a decision.
- Take photo
- Resolve scale and target
- Automatic measurement
- Return report-ready quantity
This is the shift: the useful output is not "there is a crack." It is "the crack is this wide, with this uncertainty, and this is the next action."
Product
Falcon Eyes productizes measurement AI for industrial teams.
The company is not only a product landing page and not only an academic project. Falcon Eyes is a research + product company: open research creates trust, product workflows make the measurement usable in the field.
Image to physical quantity
Width, diameter, area, spacing, count, and confidence returned as auditable inspection atoms.
Mobile-ready workflows
Designed around the camera the inspector already carries, with scale checks and failure reasons.
Vertical measurement heads
Cracks, pipes, rooms, equipment surfaces, fasteners, rebar, and customer-specific defects.
Research
GaugeAnything is the technical proof behind the company.
Foundation models tell you what and where. GaugeAnything asks the next industrial question: how many millimeters, how many instances, and which condition grade.
Research highlights
- Training-free measurement pipeline.
- No customer-specific finetuning required for the core claim.
- Mobile-ready field capture path.
- Generalizes across cracks, defects, counts, parts, and dynamic scenes.
- Audited limitations: counting and prompt vocabulary gaps are reported, not hidden.
Why now
Vision AI is ready for the physical last mile.
Segmenting, locating, and counting are useful. The next adoption barrier is measurement: physical units, uncertainty, and a clear answer an engineer can use.
Crack width
Mask for where, signal for how wide. Width profiles convert a visible crack into a measured quantity.
Defect diameter
Sharp industrial defects use diameter and area, not a generic segmentation score.
Known-object scale
Real-photo scale recovery reaches 1.74% mean relative error on coin scenes.
Moving-camera scenes
Dynamic RGB-D tests show metric signal can survive motion when object gating is correct.
Open Source
GaugeAnything Core is public, auditable, and forkable.
We keep the core visible because industrial measurement software has to earn trust. Code, weights, benchmark protocols, result logs, and limitation audits are linked directly instead of hidden behind a sales form.
Hugging Face Weights
falcons-eyes
GaugeAnything
0Stars
0Forks
1+Contributors
Apache-2.0Core license
Live GitHub stats update in the browser when available.
Benchmark
Benchmark the workflow people actually need.
GaugeBench tracks whether a method can move from pixels to physical evidence. The benchmark is not only "did it see the object?" but "did it produce the quantity the workflow depends on?"
| Method | Training | Mobile | Open | Inspection output |
|---|---|---|---|---|
| Manual CAD workflow | Human labor | No | No | Accurate, but slow and hard to scale. |
| Segmentation foundation model | No task finetuning | Partial | Varies | Masks and regions, but no physical units. |
| Count-only model | Often required | Partial | Varies | Counts, but not width, spacing, scale, or severity. |
| GaugeAnything Core | Training-free core | Yes | Yes | Millimeters, counts, scale, confidence, and caveats. |
Contact
Bring one image, one object, and one measurement rule.
Best first pilots: crack width, pipe diameter, room dimensions, equipment surface defects, fastener count, known-object scale, and RGB-D object dimensions.