Every frame, turned into insight.

Frame is Entira's video analytics platform for factories. Capture data from your cameras, label it, train your own models and run them on your own server, with alerts your teams can act on.

What Frame watches for you

Each use case below is a model trained on plant footage and a rule written in plain language. Add your own the same way.

Process

Yarn pull length by hand swing

Frame counts each hand swing of the operator and converts it to length using a per-swing value calibrated against a measured pull. A 5 m pull is confirmed without a tape.

Safety

PPE: helmet, mask, gloves, respirator

Head, face and each hand are checked separately. Stricter rules apply to a person working at an open machine than to someone walking past.

Process

Tow on floor

Material lying on the floor beyond the allowed time raises one alert that keeps its duration until the floor is clear.

Safety

Machine shutter left open

Every shutter on a machine is watched. Open for too long with nobody in the work zone means an alert with time-lapse proof.

Safety

Work at height

People on tanker tops and loading platforms are detected and checked for a safety harness.

Macro lens

Wood chip size after the chipper

Each chip is outlined and measured on the belt, giving a live split of fines, accepted and oversize.

Macro lens

Thread counting in fabric

A macro lens camera resolves single threads. Frame counts warp and weft in a measured window and reports ends and picks per inch.

Process

Dryer left unattended

A shift timeline per dryer shows attended and unattended periods. Click a red block to play the proof.

Macro lens

Coal size on the feeder

Lump size distribution is measured continuously, so oversize coal is seen before it reaches the mill.

Macro lens

Black particles in fibre

Dark specks in the fibre sheet are found, sized and counted per lot against your limit.

Safety

Two-person cutting procedure

When someone cuts a wrap, a second person must be spraying water on the cut. Frame checks both are present.

Gate

Number plate reading at the gate

Vehicle, plate and text are logged with photos for every entry and exit. The sharpest view of each plate is chosen and frames vote on the reading.

Safety

Obstruction in a pathway

Hoses, trolleys or material left in a marked walkway are flagged. People walking through never trigger it.

Footage on this page is stylised and all figures shown are simulated for illustration.

From raw camera feed to a deployed model, without leaving the app

Six steps, one application. Your team does the data preparation, training and deployment in-house.

Capture Capture profiles
Capture profileRunning
Camera
Cutter line B, Back
Mode
Detect on the fly, nothing stored unless triggered
Save a frame when
person presentevery 10 smodel unsure
Crop
People, with 20 % buffer, full resolution
Skip
Near-duplicate frames
Today's quota
1,284 of 2,000 frames
tow_on_floor
person 0.94, press Enter
Classes
  • tow_on_floor1
  • person2
  • shutter_open3
Auto-pilot

Suggestions come from your own model and improve as you label. 1,406 of 2,000 done.

Training run 14Epoch 46 of 60
Accuracy on held-back imagesLoss
91.4mAP50
94.1Precision
88.7Recall
Dataset v007
  • Train 1,612
  • Validate 204
  • Test 198
Upload a modelAlready have one? Drop the file here to register it.
Model registry41 pipelines healthy
Draft
Tow detector v8trained today
Candidate
Tow detector v7passed evaluation
Production
Tow detector v6on 14 cameras
Shutter v3on 21 cameras
Retired
Tow detector v5roll back any time
GPU
62 %
Memory
5 of 24 GB
Latency
0.9 s
Rule builderVersion 3

When tow on floor is inside Floor zone B for 10 minutes during any shift, raise a high alert with time-lapse video proof and notify Shift in-charge.

when:
  class_present: tow_on_floor
  roi: floor_zone_b
  for: 10m
then:
  alert: high
  proof: timelapse
  notify: shift_incharge
Simulator

Replayed on last week's footage, this rule would have raised

14 alerts
All alerts, Shift B3 open
StartedCameraAlertDurationProof
14:52Cutter line B, BackTow on floor18 min +Video
14:31Stretch roller 5, SideShutter open39 min +Time-lapse
14:07Stretch roller 2, FrontNo air respirator2 minPhoto
13:40Dryer 4Dryer unattended12 minTime-lapse
13:12Loading bayNo harness1 minPhoto
Sent toEmailTeamsWhatsAppSMSSlackExcelPDF
  1. 1

    Capture the right data

    Set a capture profile per camera: what to save, when, at what resolution and how to crop. Frame can detect on the fly and keep only the frames worth labelling, so disks do not fill with footage nobody needs.

  2. 2

    Label with a model beside you

    Boxes, polygons, keypoints and video clips in one labeller. Your model suggests labels, learns from every frame you accept, and auto-pilot takes over the easy ones. Import existing datasets as they are.

  3. 3

    Train, or upload a model

    Freeze a dataset version and press Train. You get an evaluation on images the model never saw, per class and per camera. Already have a model? Upload it and it joins the same registry.

  4. 4

    Deploy to cameras in a click

    Promote a model from draft to production and every camera using it reloads without a restart. One process owns the GPU, each camera has its own pipeline, and roll-back is one click.

  5. 5

    Write rules in plain language

    Pick the class, the zone, the waiting time and the shift. Test the rule on stored footage in the simulator before it goes live. Complex procedures are added as programmed rules.

  6. 6

    Alert, prove, report

    One alert per continuous violation, with photo, clip or time-lapse proof. Send it to email, Teams, WhatsApp, SMS or Slack. Summaries by plant, line, shift and use case export to Excel and PDF.

Your cameras. Your server. Your models.

Frame is installed inside your network. Data preparation, training and inference all happen on your hardware, and the models you train belong to you.

Nothing goes to the cloud

Video is decoded, analysed and stored on your server. Frame works on closed plant networks, including sites where model downloads are blocked.

Licence-clean models

Built only on open-source models with commercial-friendly licences. No per-model royalties and no surprises in an audit.

Windows or Linux

Ships as a protected build or a container. Runs on a single GPU server and scales by adding cameras.

Works with your cameras

Any IP camera with an RTSP stream, at its native resolution. Macro lens cameras for measurement use cases.

Access by role

Admin, annotator and viewer roles, with your organisation's Microsoft sign-in. Every change is logged.

Storage under control

Set how long frames and proofs are kept. Protected alerts are never removed.

Live view

See what each camera's model sees right now: boxes, zones and a timer for every rule.

Two ways to license Frame

Both include every feature on this page. Pricing depends on the number of cameras, so we quote for your site.

Perpetual

Buy once, run it for as long as you like.

  • One-time licence per server, sized by camera count
  • All use cases and the full training platform
  • Version, bug-fix and security updates
  • Add cameras when you need them
Contact for price

Subscription

Start small and pay as you grow.

  • Recurring licence, sized by camera count
  • All use cases and the full training platform
  • Updates and support included for the term
  • Scale cameras up or down at renewal
Contact for price

Show us a camera. We will show you what Frame sees.

Tell us about your plant and the problem you want to watch. We reply with a plan and a price.

sales@entira.in