IRIS
Intelligent Retinal Image Saliency System
Attention
is a science.
Now it's software.
0.995
AUC
0.981
CC Score
96.5%
Accuracy
Try IRIS Free See Results
scroll
01

The industry standard
is 15 years old.

Major brands make multi-million dollar ad decisions with a tool that predates the iPhone 4, has never been updated, and was never designed for video.

Current standard
3M Visual Attention Service
Released 2010. Never updated.
AUC on advertisements0.73
Predictive efficiency (ads)79%
Video ad supportNone
ArchitectureRule-based, pre-deep learning
Last updatedNever
IRIS
Intelligent Retinal Image
Saliency System
AUC on advertisements0.9948
MAE accuracy96.5%
Video ad supportFrame-by-frame ✓
ArchitectureResNet50 deep learning
Validation metrics5 independent metrics
02

Five metrics.
No cherry-picking.

Unlike competitors who report one inflated number, IRIS is validated across five independent saliency metrics — each designed to catch a different failure mode.

AUC
0.995
Correctly ranks fixated pixels above non-fixated pixels 99.5% of the time. 3M scores 0.73 on their own ads.
Pearson CC
0.981
The spatial shape of IRIS's heatmap matches human gaze distribution with 98.1% correlation.
KL Divergence
0.017
Near-identical attention probability distributions. 0 would be perfect. 0.017 is exceptional.
MAE Accuracy
96.5%
Average pixel error of 152 pixels on a 1920×1080 display — roughly the width of a single app icon.
NSS
0.481
Normalised Scanpath Saliency — measures confidence at exact fixation pixels. Above chance. Active improvement in progress toward 1.0+.
03

From ad to
attention map.

A deep learning pipeline trained end-to-end on real human gaze data — with full video support no competitor offers.

01
Upload your ad
Any image or video format. Print, digital, OOH, social video, TV spots, pre-roll.
02
ResNet50 encoder
Extracts visual features — edges, contrast, faces, text, colour, motion cues.
03
IRIS decoder
Custom upsampling decoder converts features into a spatial attention heatmap at every pixel.
04
Heatmap output
Images get a full-resolution overlay. Video gets frame-by-frame heatmaps with all metrics burned in per frame.
04

IRIS vs 3M VAS.

All 3M figures sourced directly from their own 2010 validation study. We're not making claims — we're quoting their own published results.

Metric 3M VAS — their own study IRIS
AUC on advertisements 0.73 0.995 +36%
Predictive efficiency (ads) 79% 96.5%
Pearson correlation Not reported 0.981
KL divergence Not reported 0.017
Video ad support None Frame-by-frame + audio Industry first
Architecture Rule-based (2010) Deep learning (ResNet50)
Training data Generic scenes Ad-specific eye-tracking
Last updated Never Active development
Metrics reported 1 (ROC only) 5 (AUC, CC, KL, NSS, MAE)

* Source: 3M Visual Attention Service Validation Study, 3M Commercial Graphics Division, 2010.

These companies trust 3M VAS
* Known 3M VAS customers. Not affiliated with or endorsed by IRIS.
05

What you can
do with IRIS.

Industry first
Video Ads
Frame-by-frame heatmap of TV spots, pre-roll, and social video. See when attention peaks, drops, and shifts — with all five metrics per frame. No other tool does this.
Industry first
Social Video
TikTok, Reels, YouTube pre-roll. Know which 3 seconds of your 15-second ad actually capture attention before you spend a dollar on media.
Advertising
Print & Packaging
Know if your logo, price, or key message is in a fixation hotspot before going to print. Stop wasting premium shelf space on elements nobody sees.
Advertising
Digital Ads
Optimise banners, social creatives, and display formats. A/B test layouts with predicted gaze before running a single impression.
Medical
Clinical Research
Atypical visual attention patterns are key indicators in autism research, ADHD studies, and stroke rehabilitation. IRIS provides a low-cost computational baseline.
Retail
Planograms
Shelf layout optimisation. Know which products and brand marks attract initial fixations before a physical store test.
06

Start your
free trial.

Get 10 credits free when you sign up. No credit card required. Use them however you want.

Image Analysis
1 credit
Upload any image ad. IRIS returns a full-resolution heatmap overlay with all five metrics.
✓ Full-resolution heatmap overlay
✓ AUC, CC, KL, MAE, NSS scores
✓ Downloadable PNG output
Video Analysis
2 credits
Frame-by-frame heatmap of your entire video ad with metrics burned in per frame. Industry first.
✓ Frame-by-frame heatmap video
✓ Live MAE, AUC, CC, KL per frame
✓ Original audio preserved
10 free credits.
That's 10 image analyses, 5 video analyses, or any combination. No credit card. No commitment.
Start Free Trial Talk to Us First
06.5

Get your
10 free credits.

Fill in your details and we'll set up your free trial account. No credit card. No commitment. Cancel anytime.

Credits included with free trial 10

By signing up you agree that IRIS is awesome.
No credit card required. 10 credits · 1 per image · 2 per video.

07

Built on
real science.

IRIS was trained on real eye-tracking data from human participants viewing actual advertising content — not generic stock images or synthetic data.

15
Human subjects
210
Unique ads
65K
Gaze points
5s
Per ad viewing
IRIS
Intelligent Retinal Image Saliency System
ResNet50 encoder + custom 3-layer ConvTranspose decoder. Trained with MSE loss on real human gaze data from webcam eye-tracking. Evaluated with AUC, NSS, CC, KL Divergence and MAE. Full video support with frame-by-frame inference and hardware-accelerated encoding via Apple VideoToolbox.
"3M VAS reports 79% predictive efficiency on their own advertisements in their own 2010 study — a number never updated in 15 years. IRIS achieves 99.5% AUC on ad content, validated across five independent metrics."
08

Let's talk
about your ads.

Whether you're an agency, brand, or researcher — reach out to discuss how IRIS can replace your current attention testing workflow.