Labels beside the action.
Evaluate names and addresses on parcels while checking that hands, objects and task instructions remain useful.
PII ZERO / PHYSICAL AILOCAL PROCESSING. REAL-WORLD DATA.
VIDEO PII REDACTION FOR ROBOTICS DATA
Keep the action.
Redact the PII.
On-prem redaction for first-person video
collected to train robots.
Stock footage · Processed with PII Zero
On-prem processingFaces, text & platesConfigurable redactionReviewable outputs
THE WORLD IS YOUR TRAINING ENVIRONMENT
Redact faces, plates
and personal text.
Detect identifying regions, track them across frames and apply blur or solid masks. Process footage on your infrastructure before annotation, sharing or training, with a policy tailored to your collection.
Designed around robotics workflows ↗01 / SEE WHAT CHANGES
How we evaluate ↗OriginalPII Zero output
Personal information goes beyond faces.
DRAG TO COMPARE ↔Find identifying text in the details: names on badges, readable messages and labels in the workspace.
OriginalPII Zero output
Sensitive text in the workspace
DRAG TO COMPARE ↔Mask recognized text on a phone while leaving the surrounding scene visible.
OriginalPII Zero output
Identifiers beyond the workbench
DRAG TO COMPARE ↔Apply plate redaction to footage captured around vehicles and public spaces.
Before-and-after demonstrations use stock footage. Media details · Coverage and evaluation
02 / FIRST-PERSON DATA COLLECTION
Explore data workflows ↗The task is the subject.
PII comes along with it.
Head-mounted and chest-mounted recordings capture human demonstrations at close range. Build the evaluation around the actions, viewpoints and identifiers in your collection.
People in the background.
Check faces, badges and personal paperwork during cooking, cleaning and assembly tasks.
Text inside the frame.
Evaluate readable messages and identifying text on phones and monitors, including brief or oblique views.
03 / VIDEO REDACTION COVERAGE
Full capability details ↗Know what your
redaction policy covers.
Visual detection
Faces & license plates
- Face regions
- Vehicle license-plate regions
Text categories · OCR required
Names & contact details
- Personal names
- Phone numbers
- Email addresses
Text categories · OCR required
Financial identifiers
- Credit-card numbers
- US bank-account numbers
- IBANs
Text categories · OCR required
US identity identifiers
- Social Security numbers
- Driver-license numbers
- Passport numbers
- Individual Taxpayer Identification Numbers (ITINs)
Optional configuration · OCR required
Extended text categories
- Locations
- Dates and times
- IP addresses
- Nationality, religious and political-group references
- US DEA registration numbers
- URLs
Optional screen-text policy
Text inside screens
- Recognized text inside detected phones
- Recognized text inside detected laptops
- Recognized text inside detected TVs / monitors
Text categories describe the configured English-language detection scope, not independently validated performance for every category. Coverage depends on readability, enabled configuration and capture conditions. Extended categories need additional review.
04 / QUALITY & PERFORMANCE
Measurement details ↗Measure recall.
Control unnecessary masking.
Set acceptance criteria on your footage, with independent labels for detection quality and a specified workload for speed and cost.
RECALL01
How much PII is found?
The share of labeled sensitive regions detected, reported separately for faces, plates and text.
PRECISION02
How many detections are correct?
The share of detections that match labeled sensitive regions, under the agreed matching rules.
MASK COVERAGE03
What remains visible?
Sensitive-region area covered, unnecessary scene masking and the effect on hands, objects and useful task text.
EXPOSURE DURATION04
How long is PII exposed?
Missed exposure time and the longest consecutive interval that identifying information remains visible.
THROUGHPUT05
How much video can you process?
Video hours per wall-clock hour, including decoding, detection, rendering and output verification.
TOTAL COST06
What does each video hour cost?
Software, compute, storage and human review effort for your resolution, categories and hardware.
These are pilot evaluation measures, not published accuracy or capacity guarantees. See the evidence page for the results measured so far.
05 / ON-PREM PROCESSING
Your infrastructure.
Your policies.
Your data stays with you.
Process footage where it already lives. Set the redaction policy, keep control of source files and put a review step before annotation, sharing or training.
Explore on-prem deployment ↗06 / PRICING
Know your cost
per video hour.
Scope pricing around your collection volume, redaction categories and deployment. Include compute and review effort when comparing the cost of processing a dataset.
Does our footage leave our infrastructure?
The redaction pipeline can run on your infrastructure. We scope model provisioning, network restrictions, storage and access with your team before deployment.
What information can PII Zero redact?
Faces, license plates and selected personal information in visible text. An optional screen-text policy masks recognized text inside detected devices. Categories and coverage are evaluated on your footage.
Can we use it for first-person robotics data?
First-person human demonstrations are our target workflow. We start with your camera setup, tasks and privacy policy, then evaluate coverage and the effect of masks on useful task information.
How do we get started?
Book a 15-minute introduction. We will discuss your collection, deployment requirements and a representative pilot. No footage upload is needed for the first conversation.
LET’S TALK VIDEO REDACTION
Discuss your video-redaction requirements.
Define your PII categories, processing volume and deployment. Talk through evaluation and pricing with our team.
15-minute introduction. No footage upload required.