Skip to content
DEEP VISION MODULE

Count and analyze your plans with precision

How many light fixtures on this plan? Sprinklers, outlets, windows? Deep Vision analyzes wide images and dense plans to deliver a near-exact count — and marks every detected element directly on the image, for immediate visual verification. One run analyzes a single page, up to 50 MB.

  • Near-exact counting
  • Every detection marked and verifiable
  • Natural-language analysis
HOW IT WORKS

Every detection is verifiable, one by one

Deep Vision never hands you a bare number: it shows its work, step by step, on the animated plan alongside.

Deep VisionIn-depth analysis
Sweeping by tile…0 detections

Breakdown by type

  • Light fixtures0
  • Double outlets0
  • Detectors0
Total detected0
  1. 1You describe what to countThe target is described in plain language — “the double outlets on the ground floor”. The model identifies it and points at one clean instance on the plan; that is where its role as a counter ends.
  2. 2A fingerprint is cutThe plan is rendered full-page and its lighting flattened, then a fingerprint of the symbol is cut around the nominated instance, trimmed tight to its own strokes.
  3. 3Deterministic censusThe fingerprint is matched across the whole page by correlation, with an ink-mass gate to reject false positives. No model call per element: on a repetitive drawing, arithmetic is not an AI's job, and this path answers in seconds.
  4. 4Announced fallback, verification and exportIf the page will not read that way, the run falls back to a sweep of overlapping tiles — slower and dearer — and tells you why it changed plan. Every kept element is marked on the image, numbered when the page holds 300 or fewer, and the whole set exports to CSV.

Measurements calibrated from the plan

When the plan already carries dimension text (e.g. 3500 mm, 12'-6"), Deep Vision uses it to calculate real-world distances between two points — not just an element count.

EXAMPLE RESULT

The count the animated plan above produces

These figures come from the example plan shown above — an illustrative example, not a guaranteed product average. The real result depends on the density and quality of your document.

0
Light fixtures detected (example)
0
Double outlets detected (example)
0
Detectors detected (example)
0
Total detected on this plan (example)
THE STAKES

A miscount costs material and margin

On a dense plan, a manual count misses elements or double-counts at the seam between sheets. The gap gets paid for in over-ordered material, job-site surprises — or margin lost on a bid that was already priced tight.

Manual takeoff

  • Tedious recounting on every large plan
  • Elements missed in dense areas or at sheet edges
  • No visual proof if the number gets challenged
  • Verification time delays the bid submission

With Deep Vision

  • A near-exact count, generated automatically instead of recounted by hand
  • The plan is tiled and analyzed in depth, with no blind spots
  • Every element is marked on the image — the proof travels with the number
  • A reliable starting point for the estimate, from the first pass
SUPPORTED DOCUMENTS

Honest handling, based on the source

Deep Vision profiles every document before analyzing it: zoom is capped to what the source can actually support, never inventing detail that isn't there — vector plans exploited at full precision, scanned plans capped to the scan's own resolution, wide-format images tiled before analysis.

Product one-pager

The one-pager — read it here, or take it with you

Read the full sheet without downloading anything, fullscreen if you prefer. The PDF stays available to share internally.

Inside this sheet

  • 1Near-exact counting
  • 2Every detection marked
  • 3Large formats, dense plans
  • 4You describe it, it counts
  • 5Calibrated measurements
  • 6Exportable proof
Deep Vision

Ready to count your plans with precision?

A count you can't verify is just an opinion: put Deep Vision in your estimating team's hands, with every detection marked and every number verifiable. Hilo Tech installs and runs your instance.