Construction Site Photo Datasets for Progress Monitoring AI

This page covers the photo-specific spec. For the full project record bundle (RFIs, submittals, schedules, cost), see construction project datasets; for continuous footage, see construction site video datasets.

Why public construction image sets rarely fit progress tracking

Public construction datasets mostly target worker safety, not installed-work state. One published site dataset holds 1,214 images from four static cameras, labeled for standing versus leaning personnel [1]. Community collections on Roboflow Universe are dominated by hardhat, vest and person classes, and their licenses vary set by set [3].

The scarcity shows up in methods. One worker-detection study trained on 1,129 generic MIT Places scene images and tested on 333 real site images, reaching roughly 70% accuracy [2]. Progress models face a harder version of that gap: they must tell a framed wall from a sheathed one, or rough-in from trim-out, on a specific floor at a specific date. Few public datasets pair that visual state with the planned activity it should match.

What a progress-ready photo record contains

A usable record links each image to time, place, trade and the planned work at that location. Without the schedule join, you have a scene-recognition set, not a progress set. The core fields:

• Capture time: EXIF DateTimeOriginal plus timezone offset, checked against upload time in the photo platform.
• Location on plan: sheet number, level, grid line or room ID, and pin coordinates if the app (Procore, Fieldwire, OpenSpace-style 360 capture or a time-lapse camera) stores them.
• Trade and element: CSI MasterFormat division, element type (slab, column, duct run, drywall) and, where available, a BIM element GUID.
• Schedule link: activity ID and planned start and finish from the Primavera P6 or Microsoft Project baseline, plus the update in force on the capture date.
• Daily log entry: crew counts, work performed and weather from the superintendent's report for that day.
• Capture mode: handheld, fixed time-lapse, 360 walk or drone, because each has different framing and repeat-visit behavior.

EXIF also carries GPS and device serials; decide what to keep using the guidance in EXIF metadata in image training data.

Request template for a progress photo dataset

The fastest way to get comparable answers from suppliers is to send the same structured request to each one. Adapt the fields below to your model target.

Illustrative example: invented to show structure; it does not describe an available dataset.

An illustrative joined record might look like this:

Illustrative example: invented to show structure; it does not describe an available dataset.

{
  "image_id": "img_000412",
  "captured_at": "2025-03-14T10:22:05-05:00",
  "capture_mode": "fixed_timelapse",
  "level": "L3",
  "grid": "C-4",
  "trade": "09 21 16 Gypsum Board Assemblies",
  "activity_id": "A3420",
  "planned_start": "2025-03-10",
  "planned_finish": "2025-03-21",
  "daily_log_ref": "DL-2025-03-14",
  "label_percent_complete": 40,
  "faces_blurred": true
}

Ownership and licensing checks before you train

Photo ownership on a project is typically set by contract, not by who held the camera. Owner-GC agreements may assign project documentation, including photos, to the owner, while subcontractor and third-party time-lapse vendor terms may reserve rights of their own. Confirm in writing who owns the archive, whether the supplying firm may license it, and whether owner consent is required.

Check these points per dataset:

1. Which contract governs project photos, and does it allow use beyond the project.
2. Whether a camera vendor's terms restrict export or secondary use.
3. Whether images show tenant spaces, artwork, logos or neighboring property that need releases (see model and property releases for AI training).
4. Whether the license names training, evaluation and model distribution separately, and states term and delivery.

For general image terms, compare against image licensing for AI training.

Worker privacy on jobsite photos

Workers appear in most progress photos, so define face and badge handling before delivery. As of October 2026, Illinois BIPA covers scans of face geometry and excludes photographs themselves, though any face-recognition processing of those photos can create consent and retention duties; the 2024 amendment (SB 2979) treats repeated collection of the same biometric from the same person by the same method as one violation [4]. Texas bars capturing a record of face geometry for a commercial purpose without notice and consent [5].

Blurring also limits model risk: image generative models have been shown to memorize and regenerate individual training photos, including people [6]. Large image collections also leak personal data through metadata that filtering misses [7]. Ask suppliers to state the blur method, the classes masked (faces, badge text, license plates, phone screens) and the sample-check results.

This page is general information, not legal advice. Confirm requirements with counsel for your jurisdiction and use case.

Common failure modes in progress photo data

Most progress datasets fail on alignment, not image quality. Watch for these:

• Clock drift: camera clocks off by hours or set to the wrong timezone, breaking the join to daily logs.
• Stale schedules: photos matched to the baseline instead of the update in force, so "behind" labels are wrong.
• Coverage gaps: handheld photos cluster on problem areas and skip routine progress, biasing percent-complete estimates.
• Leakage across splits: the same fixed camera view in train and test inflates accuracy; split by project or camera, not by image.
• Viewpoint drift: time-lapse cameras get bumped or relocated, so a "change" is really a new angle.

Construction punch-list and QA photos solve a different problem and are covered in construction QA and punch-list photos; safety detection is in construction safety image datasets.

How SourceX approaches construction photo requests

SourceX sources operational datasets from US companies on request and manages the commercial process, including licensing and ongoing purchases. Construction photo categories are not inventory, and a request does not guarantee a match. You describe the data you need; SourceX looks for US businesses that hold it, and every release is approved by the supplying company.

Each dataset is rights-reviewed for ownership and consents and delivered under a license that defines records, uses, term and delivery. Personal details are removed or replaced before delivery, the method is recorded and a sample is checked, though no method is perfect. Delivery runs through private, access-controlled workflows only after an executed agreement and supplier approval. Start a brief on the SourceX buyers page, or browse the image data hub, construction buyers and image and inspection photo licensing.

Request construction progress photos for your model

If your progress model needs time-stamped jobsite photos tied to schedules and daily logs, describe the building types, phases, capture modes and metadata you need. SourceX assesses data and licensing permissions with supplying US companies, and nothing is contracted until a supplier agrees. Describe your construction photo requirements.