GIS-ready pavement marking data, from frame to schema.
This is not generic labeling. Reviewers choose the best frame, geolocate every point against satellite context, classify stripings and markings, rate condition, and hand off cleaned geometry that production GIS teams can actually use.
414
sample frames
Video task frame count used for production proof.
1.5
sample miles
Corridor length tracked in the task report.
2.67
hours per mile
Observed task progress for the sample workflow.
509
delivery features
Records in the local final SAM production sample.
Active workflow proof
Lane stripings get two points
Linear stripings are labeled at the beginning and end of the feature so the GIS team can build continuous LineString geometry later.
Production progress
The video shows frames, miles, completion status, and hours per mile tracked in a task report.
Label taxonomy setup
The task constructor includes dozens of SAM striping and marking classes before review begins.
Image plus map review
Reviewers work with street-level imagery and a synchronized satellite map to place assets precisely.
Workflow control
Active QA decision
Lane line gap
Confidence
99.1%
Accuracy gate
>98% spatial QA
Geometry
Polyline
Category
Longitudinal
Issue
Broken centerline segment needed line continuity review.
Action
Reviewer snapped vertices, verified direction, and marked export ready.
Schema
marking_type, condition, side, direction, review_flag
Production queue
Click to sync
98%+
accuracy target
Structured review before delivery.
95%+
IAA target
Peer agreement on label consistency.
<2%
rejection target
Lead QA threshold before sign-off.
24h
rework window
Flagged items return to review quickly.
Who feels the bottleneck
Collection is finished. The delivery risk starts when extraction falls behind.
Mapping companies, mobile-mapping primes, AEC teams, and infrastructure data groups need production capacity that can absorb surge volume, clean the GIS layer, and deliver in the schema their client expects.
VP Field Operations
Collection is done, but data is not ready fast enough.
Director of Data Operations
Production backlog, inconsistent output, and staffing pressure.
QA/QC Manager
Accuracy, false positives, false negatives, and reviewer consistency.
GIS Manager
Deliverables need cleanup before they are usable.
ML Lead
Model performance depends on better training and evaluation data.
COO / VP Ops
Cost, quality, and turnaround pressure at scale.
QA and delivery system
Tight QA, then GIS cleanup, then delivery.
The operating model is not just "review labels." It is a staged QA lane with rework, escalation, schema checks, and final GIS packaging.
01
Ingest field collection
02
CVAT geolocation
03
Class and condition
04
QGIS linework
05
Schema delivery
Stage 1
CVAT checklistAnnotator self-check
Each annotator checks geometry, label taxonomy, condition, and required fields before submission.
Stage 2
95%+ IAAPeer review
A second reviewer checks label accuracy, consistency, edge cases, false positives, and false negatives.
Stage 3
<2% rejectionLead QA review
QA lead reviews flagged items, validates acceptance criteria, and confirms schema readiness.
Stage 4
Pilot sign-offClient sample review
New taxonomies can include a client review loop before production batches are signed off.
QA rules we can defend
Schema-ready output
The extraction layer can be delivered as FileGDB, shapefile, GeoJSON, CSV, PostGIS load, or the client schema from the bid. Required attributes, lane references, source frame, condition, direction, and reviewer notes can all be enforced before export.
FileGDB / Shapefile
GeoJSON / PostGIS
QA notes
Training data
What the SAM source material proves
The task is visible, measurable, and already production-shaped.
The source folders include the SAM training brief, sample images, the full workflow video, and a final GeoJSON production sample. This page now surfaces that evidence without forcing a 367 MB video into the first load.
Roadway asset inventory
GIS-ready SAM data supports this downstream use case with class, condition, geometry, and traceability intact.
Pavement marking condition
GIS-ready SAM data supports this downstream use case with class, condition, geometry, and traceability intact.
Traffic safety analysis
GIS-ready SAM data supports this downstream use case with class, condition, geometry, and traceability intact.
Mobility planning
GIS-ready SAM data supports this downstream use case with class, condition, geometry, and traceability intact.
Digital twin mapping
GIS-ready SAM data supports this downstream use case with class, condition, geometry, and traceability intact.
ML training data
GIS-ready SAM data supports this downstream use case with class, condition, geometry, and traceability intact.
Final production examples
The handoff is a named GIS package, not a vague export.
The latest examples include a combined SAM production sample, QGIS project files, conversion scripts, styling, and pilot outputs. The important part is the production lane from CVAT points to usable GIS geometry.
509
features in Final_SAM_Production_sample.geojson
327
striping records
181
marking records
12
top classes surfaced in the sample
Top classes in the production sample
Final_SAM_Production_sample.geojson
Combined production sample
MAPBOX.qgz
QGIS / Mapbox project handoff
scan_0919_BEVERLY_MA.qgz
Beverly QGIS project
cvat_xml_to_geojson_gss.py
CVAT conversion script
SAM_label_view_FIXED.qml
QGIS label styling
Geojson Converted.zip
Converted delivery package
Final_SAM_Production_sample.geojson
Local sample now included in the demo. It contains 509 features with striping and marking records in a client-ready schema.
Final_SAM_striping_points.geojson
Point layer preserving the original SAM placements before the GIS team builds final linework.
Final_SAM_striping_lines.geojson
LineString layer created from start/end points and refined with vertices to follow ground truth.
Final_SAM_markings.geojson
Point marking layer for arrows, words, symbols, crosswalk elements, and rotated marking assets.
scan_0791_GURNEE_IL_striping_lines.geojson
City-specific pilot result showing the same production pattern applied to a named corridor batch.
SAM folder production method
The offer is the full CVAT-to-QGIS production lane, not just labels.
The SAM materials show the real operating system: point placement in CVAT, class cleanup, condition rating, edge-case handling, GeoJSON conversion, QGIS line construction, symbol rotation, and final GIS handoff.
PASS 1
Pass 1: annotate and geolocate
Review the vehicle imagery, choose the clearest frame, place start and end points for stripings, place centroid points for markings, then move each point to its real-world position on the map.
PASS 2
Pass 2: classify
Replace every placeholder SAM point with a specific class. No placeholder labels should remain before export.
PASS 3
Pass 3: condition
Assign Good, Fair, or Poor based on visibility, contrast, edge clarity, paint coverage, and reflectivity where relevant.
PASS 4
Pass 4: QGIS linework
Convert CVAT output to GeoJSON, build continuous striping LineStrings, connect them with associated_id, and rotate marking symbols to match direction of travel.
Striping taxonomy
Linear features are placed with start and end points so GIS can build continuous LineString geometry.
Stripings
Marking taxonomy
Arrows, crosswalks, stop lines, words, symbols, and route shields use centroid placement and directional attributes.
Markings
Condition rating
Good
Crisp, visible, high-contrast paint with continuous coverage.
Fair
Visible but fading, with minor edge wear, gaps, or inconsistent color.
Poor
Hard to see, heavily worn, broken, low-contrast, or missing large sections.
QGIS handoff
The GIS cleanup step produces files and attributes production teams can load, inspect, and deliver.
What the QA team excludes
Where annotation usually breaks
The common pattern is simple: capture moves quickly, but extraction, QA, and schema cleanup become the bottleneck before client delivery.
Capacity that helps win bids
Proposal and operations teams need credible production capacity. This demo makes the claim concrete: human-reviewed spatial data, GIS cleanup, a defensible QA lane, and around 2,000 miles per month of current processing capacity across North America and global programs.
Pilot
Small representative sample first
Scale
Surge capacity after collection
Handoff
GIS package plus QA notes
Planning rates for bids
Per-centerline-mile rates you can carry into a DOT proposal as a subcontract line item. Final pricing follows schema, imagery quality, and QA bar review of a corridor sample.
Start with a corridor sample
Send a few miles of striping imagery or an existing marking layer.
We will review the schema, imagery quality, expected classes, exclusions, and QA bar. The response should be useful for operations, proposals, or a pilot scope.
Useful sample notes
Send a striping or marking sample
Share sample imagery, a marking layer, or a schema. We will respond with a production path, QA approach, and delivery estimate.