Off-the-shelf data · available now
Off-the-Shelf Egocentric Data Catalog
~324,800hours available now
across 18 datasets
Real, unscripted work recorded in factories, commercial sites and homes. Licensed for commercial AI training, QC'd in house and delivered in weeks.
Mono
~259,000 hrs
8 datasets
Head-mounted phone and action camera, hands in frame.
Stereo
~64,800 hrs
9 datasets
2, 4 and 6 camera rigs, with 3D hand pose on select sets.
Multi-camera
~1,000 hrs
1 dataset
Synchronized action-camera views with IMU.
Countries of collection
- Latin America
- India
- Indonesia
- Philippines
- Cambodia
- Kenya
- Uzbekistan
- China
Browse by robot task
Start from the task your robot has to do.
Pick a task to filter the catalog to the datasets tagged for it.
The catalog
Every dataset in stock, largest first.
Filter by modality, environment and task tag. Every row is an existing dataset you can request samples from.
Modality
Environment
Task tag
Showing 18 of 18 datasets
Stereo datasets
- S1~50,000 hrs
Mixed environments; breakdown by environment and task on request
- Region
- Asia
- Rig
- Head-mounted stereo
- Streams
- Stereo RGB, raw or annotated
- Task tags
- Not tagged
- S2~5,000 hrs
Factories, hotels, kitchens, cafés: sewing, ironing, carton assembly, print inspection, packing, bed making, food and coffee prep
- Region
- Cambodia
- Rig
- 2, 4 and 6 camera rigs
- Streams
- Stereo RGB, task labels, hand keypoints
- Task tags
- Assembly, Machine tending, Logistics
- S3~3,300 hrs
Commercial and industrial: construction, workshops, clothing, repair, food service, utilities, retail, automotive
- Region
- China
- Rig
- Head-mounted stereo
- Streams
- Stereo RGB, synced IMU, 3D hand pose
- Task tags
- Assembly
- S42,984 hrs
Yarn mill, textile and garment: winding, doffing, roving change, cutting, stitching, knitting, inspection
- Region
- Uzbekistan
- Rig
- Head-mounted stereo
- Streams
- Synced L+R RGB, IMU, audio
- Task tags
- Machine tending
- S5~2,000 hrs
Manufacturing across 50+ factories: appliance and LED assembly, converter lines, packaging, fastening, tool use
- Region
- Uzbekistan
- Rig
- Head stereo + 2 wrist cams
- Streams
- Depth, 200 Hz IMU, task segments, 3D hand pose
- Task tags
- Assembly, Machine tending, Logistics
- S6757 hrs
Yarn mill, blinds and woodwork: winding, combing, doffing, fitting and fastening, finishing
- Region
- Uzbekistan
- Rig
- Head-mounted stereo
- Streams
- Stereo RGB
- Task tags
- Machine tending, Assembly
- S7~550 hrs
Residential household tasks
- Region
- India
- Rig
- Head-mounted stereo, 720p
- Streams
- Stereo RGB
- Task tags
- Not tagged
- S8120 hrs
Residential: laundry, dishes, sweeping and mopping, cooking, wiping, organizing
- Region
- China
- Rig
- Head-mounted stereo
- Streams
- Stereo RGB, synced IMU, 3D hand pose
- Task tags
- Not tagged
- S9100+ hrs
Residential household tasks
- Region
- India, Philippines
- Rig
- Stereo headset, 1080p per eye
- Streams
- Stereo RGB
- Task tags
- Not tagged
Mono and multi-camera datasets
- M1~100,000 hrs
Household: cooking, food prep, kitchen and bathroom cleaning, laundry, home tasks
- Region
- Latin America led
- Rig
- Smartphone, mostly recent models
- Streams
- RGB video
- Task tags
- Not tagged
- M2~50,000 hrs
Commercial and industrial: salon and spa services, food service, retail, workshop and site work
- Region
- Latin America led
- Rig
- Smartphone, mostly recent models
- Streams
- RGB video
- Task tags
- Not tagged
- M3~50,000 hrs
Electronics and components: handling, assembly and repair
- Region
- Multi-country
- Rig
- Smartphone
- Streams
- RGB video
- Task tags
- Assembly
- M420,000 hrs
Household: cleaning, cooking, laundry, household organization
- Region
- Indonesia, Philippines, Kenya
- Rig
- Smartphone
- Streams
- RGB video
- Task tags
- Not tagged
- M520,000 hrs
Industrial and commercial: factory floor and commercial site work
- Region
- India
- Rig
- Custom head-mounted device
- Streams
- RGB video, IMU
- Task tags
- Logistics
- M612,000+ hrs
Household and commercial: home tasks, auto repair, service businesses
- Region
- India
- Rig
- Smartphone
- Streams
- RGB video
- Task tags
- Not tagged
- M76,034 hrs
Textile, garment and light manufacturing: sewing, cutting, folding, knitting, weaving, inspection, packing, woodwork
- Region
- Uzbekistan
- Rig
- Action camera, 4K
- Streams
- RGB video
- Task tags
- Machine tending, Logistics
- M8~1,000 hrs
Commercial, industrial, some household: commercial cleaning, factory, kitchen, construction, auto repair, laundry
- Region
- India
- Rig
- Smartphone, two hands in frame
- Streams
- RGB video
- Task tags
- Not tagged
- MC1Multi-camera1,000 hrs
Industrial and commercial factory floor and site work from several synchronized views
- Region
- India
- Rig
- Action cameras, multi-view
- Streams
- RGB per camera, IMU
- Task tags
- Logistics
S3 and S8 hold more inventory than listed and can grow on request. Task tags show where a dataset is tagged for Assembly, Machine tending or Logistics, and untagged sets still hold the tasks listed in their rows. 3D hand pose is on select sets, and labels can be added to any set.
Raw or labeled
Choose raw hours or add labels.
Every set can be delivered raw. Labels are applied to the hours you pick before delivery, so it stays an off-the-shelf purchase.
- Tier 1
Raw
Lowest cost per hour
Synchronized video, calibration where the rig provides it, timestamps, and IMU where captured.
- Tier 2
Pose
Adds 3D hand keypoints
Everything in Raw, plus 3D hand keypoints. 3D hand pose is already on select sets, and labels can be added to any set.
- Tier 3
Full
Adds human-reviewed segments
Everything in Pose, plus human-reviewed subtask segments with timestamps and text descriptions.
How it works
Licensed, QC'd and delivered in weeks.
License
Non-exclusive commercial AI training license. Exclusive terms are available on select stereo sets.
Quality and billing
Billed on accepted hours only. Every hour passes in-house QC and QA before it counts.
Delivery formats
Formats vary by dataset (MCAP, LeRobot, HDF5, MP4 with JSON metadata). We confirm the format per set when scoping. Raw hours are available on every set.
Beyond the shelf
Need hours beyond what is listed? We collect to your spec.
Custom collection to your specStereo data in MCAP, custom collectionOff-the-shelf stereo video
Offering 01 · Virtual tour & image data
10M+ images and 400,000+ virtual tours of real spaces.
Captured by CloudPano users in real homes, businesses and venues, and available as licensed off-the-shelf data. Every order is scoped to the industries, regions and volume you need.
- 360° images
- 10,000,000+
- Virtual tours
- 400,000+
- Industries
- Many, see below
- Pricing
- Custom quote
Drag to look around. Click the arrows to move between scenes.
Pricing
Custom quote, scoped per order.
Tour and image pricing depends on volume, industry mix and delivery scope. Reach out to a rep for a quote.
The basics
What is a virtual tour?
A virtual tour is a set of 360° photos of a real place, linked so you can move through it. Each photo, or scene, is taken from a single point and records every direction at once: floor, ceiling and all four walls. Hotspots join each scene to the next one, so the tour works like a walkable map of the building.
For a person, that is a way to visit a property remotely. For a model, it is something rarer: dense, overlapping, full-surround views of a real interior, already organized by how the space connects. The examples on this page are the same kind of tour the library holds.
- Scene
- One 360° panorama captured from a single point in a space
- Projection
- Spherical (equirectangular) images covering every direction
- Links
- Navigation hotspots that connect one scene to the next
- Tour
- The connected set of scenes for one property, walkable end to end
- Delivery scope
- Images, tour structure and available metadata, scoped and confirmed per order
Anatomy of a tour. Exact delivery contents are confirmed with your rep.
Why it matters
Why virtual tours help robotics training and 3D reconstruction.
Robots and spatial models have to work in real buildings. Tours give them real buildings at scale, captured from every direction and connected by how people move through them.
- 01
Indoor navigation and mobile robots
A tour is a graph of real places joined by walkable links. That gives navigation and embodied agents real rooms, doorways, corridors and transitions to learn layout and route-following from.
- 02
3D reconstruction and novel-view synthesis
Overlapping spherical captures of the same space are useful input for multi-view reconstruction, depth estimation and view-synthesis research, because every scene sees the whole room at once.
- 03
Scene understanding and layout
Real interiors with furniture, fixtures and clutter support room-type classification, object detection, layout estimation and floor-plan research on spaces people actually use.
- 04
Simulation and digital-twin environments
Real-world spaces can seed or validate simulated environments, helping close the gap between synthetic training rooms and the ones a robot will actually operate in.
- 05
Vision and multimodal pretraining
Ten million-plus wide-field images of real places add scene diversity to vision and vision-language pretraining beyond typical forward-facing photo collections.
- 06
Evaluation and held-out test sets
Spaces your team has never captured make honest held-out sets for checking how a model trained elsewhere generalizes to new buildings.
Coverage
Lots of industries are available.
CloudPano is used wherever a business wants to show a physical space, so the library spans far more than real estate. Examples include:
- Residential real estate
- Commercial real estate
- Hospitality and hotels
- Event venues
- Restaurants
- Retail stores
- Education and campuses
- Construction sites
- Automotive dealerships
- Healthcare facilities
- Self-storage
- Campgrounds and RV parks
Examples only. Ask your rep for available volume in a specific industry or region.
FAQ
Questions buyers ask.
- How is virtual tour and image data priced?
- Every spatial-data order is quoted individually, based on the number of tours or images, the industries and regions you need, and the delivery scope. Contact a rep and we will send a quote.
- How many egocentric hours are available now?
- About 324,800 hours across 18 datasets: about 259,000 hours of mono video in 8 datasets, about 64,800 hours of stereo video in 9 datasets, and about 1,000 hours of multi-camera video in 1 dataset.
- What are you billed for?
- Accepted hours only. Every hour passes in-house QC and QA before it counts toward your order.
- Can I see samples before buying?
- Yes. Tell us which datasets you want samples from, whether raw or labeled, and an initial volume. The tours above are live examples of the virtual tour data.
- What license comes with the egocentric datasets?
- A non-exclusive commercial AI training license. Exclusive terms are available on select stereo sets.
- Which delivery formats do you support?
- Formats vary by dataset (MCAP, LeRobot, HDF5, MP4 with JSON metadata). We confirm the format for each set when scoping. Raw hours are available on every set.
- Is 3D hand pose included?
- 3D hand pose is on select sets, and labels can be added to any set. Tell us which hours you want labeled and we apply the labels before delivery.
- What if the country or environment I need is not listed?
- We collect to spec, so gaps in the catalog can be filled with custom collection.
- How do I order?
- Tell your rep which datasets and how many hours, or which tour and image categories you need. We lock the inventory and send the order form.
Request samples
Tell us which datasets you want samples from.
Name the datasets, say whether you want them raw or labeled, and give us an initial volume. For tours and images, tell us the industries, regions and volume.
- 01 · Share the dataset IDs, raw or labeled, and an initial volume
- 02 · Review samples and metadata
- 03 · Confirm the delivery format and lock the inventory
Want to contribute footage? We are recruiting capture teams now. Apply to capture