Skip to main content
Feldy uses multiple capture techniques to understand the physical world. Photos, videos, walkthroughs, LiDAR scans, and measurement-request imagery provide different types of visual and spatial information. Feldy’s in-house vision models combine these inputs to reconstruct properties, identify relevant jobsite conditions, produce measurements, and generate project artifacts.

​
Capture methods

The capture method available depends on the project, trade, device, and enabled Feldy features.

​
Photos

Standard jobsite photos are the most common Feldy capture type. Photos help Feldy understand:
  • The overall work area
  • Visible materials and finishes
  • Damage and existing conditions
  • Building elevations
  • Rooflines and exterior transitions
  • Windows, doors, cabinets, and other objects
  • Areas included in the requested work
Photos taken from several viewpoints can also be combined to understand how different surfaces and parts of the property relate spatially.

​
Guided multi-angle capture

For exterior projects, Feldy may ask users to photograph the property from specific directions, including:
  • Front
  • Front right
  • Right side
  • Right back
  • Back
  • Back left
  • Left side
  • Left front
These overlapping views help Feldy reconstruct the exterior more accurately than a single photograph. Assigning the correct direction to each photo gives Feldy additional information about how the elevations connect.

​
Video capture

Video provides continuous visual coverage of a jobsite. Feldy can analyze video by extracting useful frames and identifying the portions that best show the property, work area, materials, or existing conditions. This means a field user can move through a space naturally while Feldy converts the video into visual information that can be understood by its vision models. Video is particularly useful for:
  • Connected rooms
  • Large work areas
  • Exterior walkthroughs
  • Complex site conditions
  • Areas that are difficult to capture in a single photograph

​
AI Job Walkthroughs

AI Job Walkthroughs combine visual capture with the field user’s spoken or written explanation. The walkthrough gives Feldy two forms of context:
  1. What the camera sees
  2. What the field user says about the work
For example, a user may point the camera at damaged siding and explain:
Replace the damaged siding on the rear elevation, but do not include the adjacent garage wall.
Feldy can use both the visual information and the instruction when generating an estimate, scope of work, or project document.

​
LiDAR capture

Supported mobile devices can use LiDAR to capture depth and spatial geometry. LiDAR helps Feldy understand:
  • Wall locations
  • Floor boundaries
  • Room dimensions
  • Openings
  • Connected spaces
  • The distance between surfaces
  • The approximate shape and scale of the captured environment
Interior LiDAR capture currently supports flooring and wall measurements for eligible accounts.

​
Measurement-order photos

Roofing and exterior organizations can attach project photos when ordering a measurement report. These photos provide additional context that may not be visible in aerial imagery, including:
  • Roof sections obscured by trees
  • Building additions
  • Dormers
  • Low-slope sections
  • Attached garages
  • Exterior projections
  • Complex roof-to-wall transitions
  • Structures that should be included or excluded
Providing clear jobsite photos helps Feldy create more complete roof and exterior models.

​
How Feldy processes captures

Feldy’s in-house vision models use several computer-vision and spatial-computing techniques to transform raw captures into structured project information. The exact processing method depends on the capture type and requested output.

​
Visual understanding

Feldy analyzes captures to identify relevant parts of the physical environment. Depending on the project, this can include:
  • Walls
  • Floors
  • Roof surfaces
  • Windows and doors
  • Cabinets
  • Exterior elevations
  • Materials
  • Visible damage
  • Building openings
  • Attached structures
  • Areas described by the user
This allows Feldy to move beyond simply storing photos and begin understanding what is shown within them.

​
Video-frame extraction

When a user records video, Feldy can extract individual frames from the recording. The system selects frames that provide useful, non-duplicative views of the work area. These frames can then be processed similarly to standard jobsite photos. This allows video to become a structured capture source rather than remaining only a recording that someone must manually review.

​
Multi-view reconstruction

When multiple images show the same property or space from different viewpoints, Feldy can compare shared visual features across them. This helps determine:
  • How surfaces connect
  • The relative position of building features
  • The shape of the property
  • Which objects appear across several views
  • How the camera moved around the structure
The result can be used to reconstruct a spatial representation of the captured environment.

​
Gaussian splatting

Feldy may use Gaussian splatting to produce a detailed visual reconstruction from overlapping images or video frames. Gaussian splatting creates a navigable representation of the captured property that preserves much of the appearance of the original scene. This can help teams:
  • Review the property from additional viewpoints
  • Understand the captured environment spatially
  • Inspect areas across several connected images
  • Preserve a visual record of the jobsite
Gaussian splatting is one part of the reconstruction process. Structured measurements are produced from the geometric and measurement layers of the property model rather than relying only on the visual splat.

​
LiDAR geometry

For supported interior scans, LiDAR provides direct depth information from the mobile device. Feldy uses this information to create structured geometry representing the room, including walls, floors, openings, and other detected features. Because LiDAR provides scale, it is especially useful for producing dimensional information from an interior capture.

​
Structured property modeling

Visual reconstruction is converted into a structured model that Feldy can use for measurements and downstream workflows. A structured model may include:
  • Roof planes
  • Exterior walls
  • Interior walls
  • Floor surfaces
  • Windows and doors
  • Openings
  • Building edges
  • Attached structures
  • Spatial dimensions
This structured representation is what allows Feldy to move from a visual capture to a usable project artifact.

​
From capture to project output

Feldy’s capture workflow can be summarized as:

​
1. Capture the property

The field team records photos, videos, walkthroughs, or 3D scans.

​
2. Understand the scene

Feldy’s vision models identify surfaces, objects, geometry, and relevant jobsite conditions.

​
3. Reconstruct the space

Multiple views, video frames, LiDAR data, and other available inputs are combined into a spatial representation.

​
4. Structure the information

Feldy converts the reconstructed scene into usable property data, such as walls, floors, roof planes, openings, and measurements.

​
5. Generate an artifact

The structured information can then power:
  • Estimates
  • Measurement reports
  • 3D property models
  • Scopes of work
  • Inspection documents
  • Project summaries
  • Proposals
  • Customer visualizations

​
Feldy’s spatial AI vision models

Captures do more than document the jobsite. They provide the inputs that power Feldy’s in-house spatial vision models. These models are designed to understand properties and physical work by combining:
  • Visual appearance
  • Spatial geometry
  • Depth information
  • Multiple viewpoints
  • Spoken and written instructions
  • Project and property context
The goal is to turn what happens in the field into structured information that can be measured, estimated, documented, and communicated.

​
Capture quality matters

The quality and completeness of the inputs affect the quality of the resulting model and project outputs. For better results:
  • Capture the complete work area
  • Use several viewpoints
  • Keep images clear and well lit
  • Include corners and transitions
  • Avoid excessive blur
  • Capture areas hidden from aerial imagery
  • Explain unusual conditions
  • Attach relevant photos to measurement orders
  • Review generated measurements and models
Feldy can automate much of the interpretation and reconstruction process, but users should still review generated measurements, estimates, and documentation before relying on them for final pricing, material ordering, or construction.