EMMA-CVPR2026 / docs /index.html
abanerj3's picture
Upload folder using huggingface_hub
b46439f verified
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>EMMA — Extracting Multiple physical parameters from Multimodal Data</title>
<meta name="description" content="EMMA: a physics-informed multimodal framework for recovering dynamical parameters from video (with optional audio and chart-image support). CVPR 2026." />
<meta property="og:title" content="EMMA: Extracting Multiple physical parameters from Multimodal Data" />
<meta property="og:description" content="CVPR 2026. Physics-informed parameter recovery from video, with optional audio and plots." />
<meta property="og:type" content="website" />
<link rel="preconnect" href="https://fonts.googleapis.com" />
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
<link href="https://fonts.googleapis.com/css2?family=Fraunces:opsz,wght@9..144,400;9..144,500;9..144,600&family=Inter:wght@400;500;600&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet" />
<!-- KaTeX for equations -->
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.16.11/dist/katex.min.css" crossorigin="anonymous" />
<script defer src="https://cdn.jsdelivr.net/npm/katex@0.16.11/dist/katex.min.js" crossorigin="anonymous"></script>
<script defer src="https://cdn.jsdelivr.net/npm/katex@0.16.11/dist/contrib/auto-render.min.js"
crossorigin="anonymous"
onload="renderMathInElement(document.body,{delimiters:[{left:'$$',right:'$$',display:true},{left:'$',right:'$',display:false}]});"></script>
<style>
:root {
--bg: #ffffff;
--bg-soft: #f5f2ea;
--paper: #fbfaf6;
--ink: #161616;
--ink-2: #2f2f33;
--ink-3: #5f5f66;
--ink-4: #8a8a92;
--rule: #e7e3d5;
--rule-strong: #c9bfa4;
--accent: #a84318;
--accent-ink: #7f3010;
}
* { box-sizing: border-box; }
html { scroll-behavior: smooth; }
body {
margin: 0;
background: var(--bg);
color: var(--ink);
font-family: "Inter", ui-sans-serif, system-ui, -apple-system, sans-serif;
font-size: 16.5px;
line-height: 1.65;
-webkit-font-smoothing: antialiased;
text-rendering: optimizeLegibility;
}
.wrap { max-width: 960px; margin: 0 auto; padding: 0 28px; }
.narrow { max-width: 760px; margin-left: auto; margin-right: auto; }
a { color: var(--accent-ink); text-decoration: none; }
a:hover { text-decoration: underline; }
/* HEADER */
header.hero {
padding: 76px 0 40px;
text-align: center;
}
.venue {
display: inline-block;
font-family: "JetBrains Mono", monospace;
font-size: 11.5px;
letter-spacing: .22em;
text-transform: uppercase;
color: var(--accent-ink);
background: rgba(168,67,24,.08);
border: 1px solid rgba(168,67,24,.22);
padding: 6px 12px;
border-radius: 999px;
margin-bottom: 22px;
}
h1.title {
font-family: "Fraunces", "Times New Roman", serif;
font-weight: 600;
font-size: clamp(28px, 3.8vw, 40px);
line-height: 1.2;
letter-spacing: -.015em;
margin: 4px auto 16px;
color: var(--ink);
max-width: 900px;
}
h1.title .acronym { font-weight: 700; }
h1.title em {
font-weight: 500;
font-style: italic;
color: var(--ink-2);
}
.tldr {
font-family: "Fraunces", serif;
font-style: italic;
font-weight: 400;
font-size: clamp(15px, 1.6vw, 17px);
color: var(--ink-3);
max-width: 700px;
margin: 0 auto 26px;
line-height: 1.5;
letter-spacing: -.005em;
}
.authors-block { margin: 8px auto 32px; max-width: 820px; }
ol.authors {
list-style: none; padding: 0; margin: 0 0 10px;
display: flex; flex-wrap: wrap; justify-content: center;
gap: 4px 24px;
font-family: "Fraunces", serif;
font-size: 18px;
color: var(--ink);
}
ol.authors li { display: inline-block; }
ol.authors a {
color: var(--ink); text-decoration: none;
border-bottom: 1px dotted var(--rule-strong);
padding-bottom: 1px;
transition: color .15s ease, border-color .15s ease;
}
ol.authors a:hover { color: var(--accent); border-color: var(--accent); }
.affil-line {
font-size: 14px;
color: var(--ink-3);
margin: 0 0 20px;
text-align: center;
}
.logo-row {
display: flex; align-items: center; justify-content: center;
gap: 28px;
margin-top: 4px;
}
.logo-row .logo-link {
display: inline-flex; align-items: center;
padding: 4px 8px;
border-radius: 6px;
transition: background .15s ease;
}
.logo-row .logo-link:hover { background: var(--bg-soft); }
.logo-row img {
height: 42px; width: auto; display: block;
opacity: .92;
}
@media (max-width: 480px) {
.logo-row { gap: 18px; }
.logo-row img { height: 34px; }
}
.actions {
display: flex; flex-wrap: wrap; justify-content: center; gap: 10px;
}
.btn {
display: inline-flex; align-items: center; gap: 8px;
padding: 10px 18px;
font-size: 14px; font-weight: 500;
background: var(--ink);
color: #fffcf5;
border: 1px solid var(--ink);
border-radius: 999px;
transition: background .15s ease, color .15s ease, border-color .15s ease;
}
.btn:hover { background: var(--accent); border-color: var(--accent); color: #fffcf5; text-decoration: none; }
.btn.ghost { background: transparent; color: var(--ink); }
.btn.ghost:hover { background: var(--ink); color: #fffcf5; border-color: var(--ink); }
.btn svg { width: 14px; height: 14px; }
hr.rule { border: none; height: 1px; background: var(--rule); margin: 0; }
/* SECTIONS */
section { padding: 64px 0; }
.kicker {
font-family: "JetBrains Mono", monospace;
font-size: 14px;
letter-spacing: .24em;
text-transform: uppercase;
color: var(--accent);
margin-bottom: 18px;
}
h2 {
font-family: "Fraunces", serif;
font-weight: 500;
font-size: clamp(34px, 5vw, 52px);
line-height: 1.1;
letter-spacing: -.015em;
margin: 0 0 24px;
color: var(--ink);
}
h2 .em { font-style: italic; color: var(--accent); font-weight: 400; }
p { margin: 0 0 14px; color: var(--ink-2); }
p:last-child { margin-bottom: 0; }
.lead { font-size: 18px; color: var(--ink-2); line-height: 1.6; max-width: 720px; }
.tagline {
font-family: "Fraunces", serif;
font-size: clamp(18px, 2vw, 22px);
color: var(--ink-3);
margin: -8px 0 16px;
letter-spacing: -.005em;
}
.tagline .em {
font-style: italic;
color: var(--accent);
font-weight: 500;
margin-right: 4px;
}
/* PITCH / ADVERTISEMENT */
#pitch { padding: 72px 0; }
.pitch-head { text-align: center; margin-bottom: 44px; }
.pitch-rows {
display: flex; flex-direction: column;
border-top: 1px solid var(--rule);
}
.pitch-row {
display: grid;
grid-template-columns: 1fr auto 1fr;
gap: 24px;
align-items: center;
padding: 22px 0;
border-bottom: 1px solid var(--rule);
}
.pitch-row .problem {
font-family: "Fraunces", serif;
font-size: 20px;
color: var(--ink);
line-height: 1.35;
}
.pitch-row .arrow {
font-family: "JetBrains Mono", monospace;
font-size: 14px;
color: var(--rule-strong);
white-space: nowrap;
}
.pitch-row .answer { font-size: 16px; color: var(--ink-2); line-height: 1.5; }
.pitch-row .answer b { color: var(--accent-ink); font-weight: 600; }
@media (max-width: 680px) {
.pitch-row { grid-template-columns: 1fr; gap: 8px; padding: 18px 0; }
.pitch-row .arrow { display: none; }
.pitch-row .problem { font-size: 18px; }
}
/* IO PANEL */
.io-panel {
display: grid;
grid-template-columns: 1fr 1fr;
gap: 1px;
background: var(--rule);
border: 1px solid var(--rule);
border-radius: 10px;
overflow: hidden;
margin-top: 40px;
}
.io-col { background: var(--paper); padding: 28px 26px; }
.io-col .label {
font-family: "JetBrains Mono", monospace;
font-size: 11.5px;
letter-spacing: .2em;
text-transform: uppercase;
color: var(--ink-3);
margin-bottom: 14px;
}
.io-col h3 {
font-family: "Fraunces", serif;
font-weight: 500;
font-size: 20px;
line-height: 1.25;
margin: 0 0 14px;
color: var(--ink);
}
.io-col ul { list-style: none; padding: 0; margin: 0; }
.io-col li {
padding: 10px 0;
font-size: 15.5px;
color: var(--ink-2);
border-top: 1px dashed var(--rule);
line-height: 1.45;
}
.io-col li:first-child { border-top: none; }
.io-col.out { background: #fff9ec; }
.io-col.out li { color: var(--ink); }
.io-col .tag {
display: inline-block;
font-family: "JetBrains Mono", monospace;
font-size: 10.5px;
letter-spacing: .1em;
text-transform: uppercase;
color: var(--accent-ink);
background: rgba(168,67,24,.08);
border: 1px solid rgba(168,67,24,.2);
border-radius: 999px;
padding: 2px 8px;
margin-left: 6px;
vertical-align: 1px;
}
.io-col .tag.muted {
color: var(--ink-3);
background: transparent;
border-color: var(--rule-strong);
}
@media (max-width: 680px) { .io-panel { grid-template-columns: 1fr; } }
/* PHYSICS / EQUATIONS */
.eq-block {
background: var(--paper);
border: 1px solid var(--rule);
border-radius: 10px;
padding: 28px 28px;
margin-top: 20px;
}
.eq-block .eq-row {
padding: 14px 0;
border-top: 1px dashed var(--rule);
}
.eq-block .eq-row:first-child { border-top: none; padding-top: 0; }
.eq-block .eq-row:last-child { padding-bottom: 0; }
.eq-block .eq-caption {
font-size: 14.5px;
color: var(--ink-3);
margin-top: 6px;
line-height: 1.5;
}
/* FIGURE */
figure.fig {
margin: 28px 0 0;
padding: 18px;
background: var(--paper);
border: 1px solid var(--rule);
border-radius: 10px;
text-align: center;
}
figure.fig img {
max-width: 100%;
height: auto;
display: block;
margin: 0 auto;
border-radius: 4px;
}
figure.fig figcaption {
margin-top: 14px;
padding: 0 8px;
font-size: 14.5px;
color: var(--ink-3);
line-height: 1.55;
text-align: left;
max-width: 820px;
margin-left: auto;
margin-right: auto;
}
figure.fig figcaption b { color: var(--ink); font-weight: 600; }
/* GALLERY */
.gallery {
display: grid;
grid-template-columns: 1fr 1fr;
gap: 18px;
margin-top: 28px;
}
.gtile {
margin: 0;
background: var(--paper);
border: 1px solid var(--rule);
border-radius: 10px;
overflow: hidden;
}
.gtile .gt-label {
display: flex; align-items: baseline; justify-content: space-between; gap: 10px;
padding: 12px 16px;
font-family: "Fraunces", serif;
font-weight: 500;
font-size: 17px;
color: var(--ink);
border-bottom: 1px solid var(--rule);
background: var(--bg-soft);
}
.gtile .gt-label span {
font-family: "JetBrains Mono", monospace;
font-weight: 400;
font-size: 11.5px;
letter-spacing: .06em;
color: var(--ink-3);
}
.gtile .gt-bodies {
display: grid;
grid-template-columns: 1fr 1fr;
gap: 1px;
background: var(--rule);
}
.gtile .gt-pane {
background: var(--paper);
position: relative;
display: flex; align-items: center; justify-content: center;
aspect-ratio: 4 / 3;
overflow: hidden;
}
.gtile .gt-pane img {
width: 100%; height: 100%; object-fit: cover; display: block;
}
.gtile .gt-cap {
position: absolute;
left: 6px; bottom: 6px;
font-family: "JetBrains Mono", monospace;
font-size: 8.5px;
letter-spacing: .1em;
text-transform: uppercase;
padding: 2px 5px;
border-radius: 3px;
background: rgba(22,22,22,.72);
color: #fffcf5;
backdrop-filter: blur(2px);
}
.gtile .gt-pane:last-child .gt-cap {
background: rgba(168,67,24,.92);
}
@media (max-width: 820px) {
.gallery { grid-template-columns: 1fr; }
.gtile .gt-label { font-size: 15.5px; }
}
@media (max-width: 420px) {
.gtile .gt-bodies { grid-template-columns: 1fr; }
}
/* VIDEO */
.yt-card {
display: block;
position: relative;
aspect-ratio: 16 / 9;
background: #000;
border: 1px solid var(--rule-strong);
border-radius: 10px;
overflow: hidden;
text-decoration: none;
margin-top: 8px;
}
.yt-card img {
width: 100%; height: 100%; object-fit: cover; display: block;
transition: transform .4s ease;
}
.yt-card:hover img { transform: scale(1.02); }
.yt-overlay {
position: absolute; inset: 0;
display: grid; place-items: center;
background: linear-gradient(180deg, rgba(0,0,0,.1) 40%, rgba(0,0,0,.5) 100%);
}
.yt-play {
width: 76px; height: 76px;
border-radius: 50%;
background: rgba(255,252,245,.95);
display: grid; place-items: center;
box-shadow: 0 10px 24px rgba(0,0,0,.35);
transition: transform .2s ease;
}
.yt-card:hover .yt-play { transform: scale(1.06); }
.yt-play::after {
content: ""; width: 0; height: 0;
border-left: 20px solid var(--ink);
border-top: 13px solid transparent;
border-bottom: 13px solid transparent;
margin-left: 5px;
}
.yt-badge {
position: absolute;
left: 18px; bottom: 18px;
font-family: "JetBrains Mono", monospace;
font-size: 11.5px;
letter-spacing: .14em;
text-transform: uppercase;
padding: 6px 12px;
border-radius: 999px;
background: rgba(22,22,22,.72);
color: #fffcf5;
backdrop-filter: blur(3px);
}
/* METHOD */
ol.stages {
list-style: none; padding: 0; margin: 22px 0 0;
counter-reset: stage;
display: grid; grid-template-columns: 1fr 1fr 1fr; gap: 16px;
}
ol.stages li {
counter-increment: stage;
background: var(--paper);
border: 1px solid var(--rule);
border-radius: 10px;
padding: 22px 20px;
}
ol.stages li::before { content: none; }
ol.stages .step-tag {
display: inline-block;
font-family: "JetBrains Mono", monospace;
font-size: 11px;
letter-spacing: .18em;
text-transform: uppercase;
color: var(--accent-ink);
background: rgba(168,67,24,.08);
border: 1px solid rgba(168,67,24,.22);
padding: 4px 9px;
border-radius: 999px;
margin-bottom: 14px;
}
ol.stages h3 {
font-family: "Fraunces", serif;
font-weight: 500;
font-size: 18px;
line-height: 1.3;
margin: 0 0 8px;
color: var(--ink);
}
ol.stages p { font-size: 15px; line-height: 1.5; margin: 0; color: var(--ink-2); }
@media (max-width: 820px) { ol.stages { grid-template-columns: 1fr; } }
/* RESULTS OVERVIEW */
.res-grid {
display: grid;
grid-template-columns: repeat(4, 1fr);
gap: 14px;
margin: 30px 0 0;
}
.res-card {
background: var(--paper);
border: 1px solid var(--rule);
border-radius: 10px;
padding: 22px 20px 20px;
}
.res-num {
font-family: "Fraunces", serif;
font-weight: 600;
font-size: 44px;
line-height: 1;
letter-spacing: -.03em;
color: var(--accent-ink);
margin-bottom: 8px;
}
.res-num span {
font-size: 26px;
color: var(--accent);
font-weight: 500;
margin-left: 2px;
}
.res-label {
font-family: "JetBrains Mono", monospace;
font-size: 11px;
letter-spacing: .14em;
text-transform: uppercase;
color: var(--ink-3);
margin-bottom: 10px;
}
.res-note {
font-size: 13.5px;
color: var(--ink-2);
line-height: 1.45;
}
@media (max-width: 820px) { .res-grid { grid-template-columns: repeat(2, 1fr); } }
@media (max-width: 480px) { .res-grid { grid-template-columns: 1fr; } }
.eval-sub {
font-family: "Fraunces", serif;
font-weight: 500;
font-size: 22px;
margin: 44px 0 0;
color: var(--ink);
letter-spacing: -.01em;
}
/* SYSTEMS */
.sys-grid {
display: grid; grid-template-columns: 1fr 1fr 1fr; gap: 16px; margin-top: 16px;
}
.sys-card {
background: var(--paper);
border: 1px solid var(--rule);
border-radius: 10px;
padding: 22px 20px;
}
.sys-card .cat {
font-family: "JetBrains Mono", monospace;
font-size: 11.5px;
letter-spacing: .18em;
text-transform: uppercase;
color: var(--ink-3);
margin-bottom: 12px;
}
.sys-card ul { list-style: none; padding: 0; margin: 0; }
.sys-card li {
font-size: 15px; padding: 7px 0;
border-top: 1px dashed var(--rule);
color: var(--ink-2);
}
.sys-card li:first-child { border-top: none; }
@media (max-width: 820px) { .sys-grid { grid-template-columns: 1fr; } }
/* CITATION */
pre.bibtex {
background: #141418;
color: #e8e1cf;
border-radius: 10px;
padding: 22px 24px;
font-family: "JetBrains Mono", monospace;
font-size: 13px;
line-height: 1.7;
overflow-x: auto;
margin: 20px 0 0;
border: 1px solid #2a2a30;
}
pre.bibtex .k { color: #e89562; }
pre.bibtex .v { color: #b8d4c6; }
.bibtex-wrap { position: relative; }
.bibtex-copy {
position: absolute;
top: 14px;
right: 14px;
display: inline-flex;
align-items: center;
gap: 6px;
padding: 6px 12px;
font-family: inherit;
font-size: 12px;
font-weight: 500;
color: #e8e1cf;
background: #2a2a30;
border: 1px solid #3a3a42;
border-radius: 6px;
cursor: pointer;
transition: background 0.15s, border-color 0.15s, color 0.15s;
}
.bibtex-copy:hover { background: #35353d; border-color: var(--accent); color: var(--accent); }
.bibtex-copy.copied { color: #b8d4c6; border-color: #b8d4c6; }
.bibtex-copy svg { width: 14px; height: 14px; }
/* FOOTER */
footer {
padding: 40px 0 56px;
border-top: 1px solid var(--rule);
color: var(--ink-3);
font-size: 14px;
}
footer .row {
display: flex; justify-content: space-between; flex-wrap: wrap; gap: 18px;
}
footer a { color: var(--ink-2); border-bottom: 1px solid var(--rule-strong); }
footer a:hover { color: var(--accent); border-color: var(--accent); text-decoration: none; }
.ack {
margin-top: 18px;
padding: 16px 20px;
background: var(--bg-soft);
border-left: 3px solid var(--accent);
border-radius: 4px;
font-size: 14px;
color: var(--ink-2);
max-width: 760px;
}
a:focus-visible, .btn:focus-visible, button:focus-visible {
outline: 2px solid var(--accent);
outline-offset: 2px;
}
</style>
</head>
<body>
<!-- HEADER -->
<header class="hero">
<div class="wrap">
<span class="venue">CVPR 2026</span>
<h1 class="title"><span class="acronym">EMMA:</span> <em>Extracting Multiple physical parameters from Multimodal Data</em></h1>
<p class="tldr">Recover the physics of a system from raw video or image, even when the hidden dynamics are unknown.</p>
<div class="authors-block">
<ol class="authors">
<li><a href="https://scholar.google.com/citations?hl=en&amp;user=mbAOSW0AAAAJ" target="_blank" rel="noopener">Farhat Shaikh</a></li>
<li><a href="https://scholar.google.com/citations?user=UAlc7tEAAAAJ&amp;hl=en" target="_blank" rel="noopener">Ayan Banerjee</a></li>
<li><a href="https://scholar.google.com/citations?user=U9bcQkMAAAAJ&amp;hl=en" target="_blank" rel="noopener">Sandeep K. S. Gupta</a></li>
</ol>
<p class="affil-line">IMPACT Lab, School of Computing &amp; Augmented Intelligence &middot; Arizona State University</p>
<div class="logo-row" aria-hidden="true">
<a href="https://impact.lab.asu.edu/" target="_blank" rel="noopener" class="logo-link" aria-label="IMPACT Lab">
<img src="impact-lab-logo.png" alt="IMPACT Lab" />
</a>
<a href="https://www.asu.edu/" target="_blank" rel="noopener" class="logo-link" aria-label="Arizona State University">
<img src="ASU-logo.png" alt="Arizona State University" />
</a>
</div>
</div>
<div class="actions">
<a class="btn" href="https://openaccess.thecvf.com/content/CVPR2026/html/Shaikh_EMMA_Extracting_Multiple_physical_parameters_from_Multimodal_Data_CVPR_2026_paper.html" target="_blank" rel="noopener">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M14 2H6a2 2 0 0 0-2 2v16a2 2 0 0 0 2 2h12a2 2 0 0 0 2-2V8z"/><polyline points="14 2 14 8 20 8"/></svg>
Paper
</a>
<a class="btn ghost" href="https://github.com/ImpactLabASU/EMMA-CVPR2026" target="_blank" rel="noopener">
<svg viewBox="0 0 24 24" fill="currentColor"><path d="M12 .5C5.73.5.5 5.73.5 12a11.5 11.5 0 0 0 7.86 10.92c.58.1.79-.25.79-.56v-2c-3.2.7-3.88-1.38-3.88-1.38-.52-1.33-1.28-1.68-1.28-1.68-1.05-.72.08-.7.08-.7 1.16.08 1.77 1.2 1.77 1.2 1.03 1.76 2.7 1.25 3.36.96.1-.75.4-1.25.73-1.54-2.55-.29-5.24-1.28-5.24-5.7 0-1.26.45-2.29 1.19-3.1-.12-.3-.52-1.48.11-3.08 0 0 .97-.31 3.18 1.18a11 11 0 0 1 5.79 0c2.2-1.5 3.17-1.18 3.17-1.18.63 1.6.23 2.78.11 3.08.74.81 1.19 1.84 1.19 3.1 0 4.43-2.7 5.4-5.26 5.69.41.36.78 1.07.78 2.16v3.2c0 .31.21.67.8.56A11.5 11.5 0 0 0 23.5 12C23.5 5.73 18.27.5 12 .5z"/></svg>
Code
</a>
<a class="btn ghost" href="#video">
<svg viewBox="0 0 24 24" fill="currentColor"><path d="M23 7l-7 5 7 5V7z"/><rect x="1" y="5" width="15" height="14" rx="2"/></svg>
Video
</a>
</div>
</div>
</header>
<hr class="rule" />
<!-- GALLERY: real footage beside EMMA's simulated trajectory -->
<section id="gallery">
<div class="wrap">
<div class="kicker">Gallery</div>
<h2>Real footage, <span class="em">simulated from recovered parameters.</span></h2>
<p class="lead">
For each system, the left panel is the input video; the right panel is a forward
simulation driven by the physical parameters EMMA recovered from that video.
</p>
<div class="gallery">
<figure class="gtile">
<figcaption class="gt-label">Pendulum <span>L = 0.86 m, &tau; = 0.045</span></figcaption>
<div class="gt-bodies">
<div class="gt-pane"><img src="media/pendulum-real.gif" alt="Real pendulum video" loading="lazy" /><span class="gt-cap">actual</span></div>
<div class="gt-pane"><img src="media/pendulum-sim.gif" alt="Pendulum simulated from recovered parameters" loading="lazy" /><span class="gt-cap">EMMA</span></div>
</div>
</figure>
<figure class="gtile">
<figcaption class="gt-label">Free fall <span>g = 9.8 m/s<sup>2</sup></span></figcaption>
<div class="gt-bodies">
<div class="gt-pane"><img src="media/free-fall-real.gif" alt="Real free fall video" loading="lazy" /><span class="gt-cap">actual</span></div>
<div class="gt-pane"><img src="media/free-fall-sim.gif" alt="Free fall simulated from recovered parameters" loading="lazy" /><span class="gt-cap">EMMA</span></div>
</div>
</figure>
<figure class="gtile">
<figcaption class="gt-label">Torricelli drainage <span>k = 0.0132 &radic;m/s</span></figcaption>
<div class="gt-bodies">
<div class="gt-pane"><img src="media/torricelli-real.gif" alt="Real draining container video" loading="lazy" /><span class="gt-cap">actual</span></div>
<div class="gt-pane"><img src="media/torricelli-sim.gif" alt="Torricelli drainage simulated from recovered parameters" loading="lazy" /><span class="gt-cap">EMMA</span></div>
</div>
</figure>
<figure class="gtile">
<figcaption class="gt-label">Sliding block <span>&alpha; = 24.7&deg;, &mu; = 0.205</span></figcaption>
<div class="gt-bodies">
<div class="gt-pane"><img src="media/sliding-block-real.gif" alt="Real sliding block video" loading="lazy" /><span class="gt-cap">actual</span></div>
<div class="gt-pane"><img src="media/sliding-block-sim.gif" alt="Sliding block simulated from recovered parameters" loading="lazy" /><span class="gt-cap">EMMA</span></div>
</div>
</figure>
<figure class="gtile">
<figcaption class="gt-label">LED decay <span>&gamma; = 0.91</span></figcaption>
<div class="gt-bodies">
<div class="gt-pane"><img src="media/led-real.gif" alt="Real LED decay video" loading="lazy" /><span class="gt-cap">actual</span></div>
<div class="gt-pane"><img src="media/led-sim.gif" alt="LED decay simulated from recovered parameters" loading="lazy" /><span class="gt-cap">EMMA</span></div>
</div>
</figure>
<figure class="gtile">
<figcaption class="gt-label">Rover <span>5 params</span></figcaption>
<div class="gt-bodies">
<div class="gt-pane"><img src="media/rover-real.gif" alt="Real differential-drive rover video" loading="lazy" /><span class="gt-cap">actual</span></div>
<div class="gt-pane"><img src="media/rover-sim.gif" alt="Rover trajectory simulated from recovered parameters" loading="lazy" /><span class="gt-cap">EMMA</span></div>
</div>
</figure>
<figure class="gtile">
<figcaption class="gt-label">Quadrotor <span>7 params</span></figcaption>
<div class="gt-bodies">
<div class="gt-pane"><img src="media/drone-real.gif" alt="Real quadrotor video" loading="lazy" /><span class="gt-cap">actual</span></div>
<div class="gt-pane"><img src="media/drone-sim.gif" alt="Quadrotor trajectory simulated from recovered parameters" loading="lazy" /><span class="gt-cap">EMMA</span></div>
</div>
</figure>
</div>
</div>
</section>
<hr class="rule" />
<!-- PITCH / ADVERTISEMENT -->
<section id="pitch">
<div class="wrap">
<div class="pitch-head">
<div class="kicker">Why EMMA</div>
<h2>Give multimodal data (video, image). <span class="em">EMMA figures out the system.</span></h2>
<p class="lead" style="margin: 14px auto 0;">
Point a camera at the system, or hand EMMA a published plot. Provide its
equation of motion. EMMA recovers the dynamical parameters, the hidden
actuation, and the coordinate frame, all inside one continuous-time model.
When audio is available, the same pipeline fuses it with video to resolve
states that video alone cannot see.
</p>
</div>
<div class="io-panel">
<div class="io-col">
<div class="label">You give EMMA</div>
<h3>A video or a chart, plus the equation of motion.</h3>
<ul>
<li><b>Video of the system</b> <span class="tag">visual input</span><br/><span style="color:var(--ink-3);font-size:14px;">raw frames, no masks, no templates</span></li>
<li><b>Chart images</b> <span class="tag">visual input</span><br/><span style="color:var(--ink-3);font-size:14px;">an alternative to video, for systems known only through published plots</span></li>
<li><b>Governing ODE form</b> <span class="tag">required</span><br/><span style="color:var(--ink-3);font-size:14px;">the parametric structure of <code>dx/dt = f(x, u; θ)</code> and the parameters you want recovered</span></li>
<li><b>Audio</b> <span class="tag muted">optional, multimodal</span><br/><span style="color:var(--ink-3);font-size:14px;">paired with video when states or actuation are hidden from the camera</span></li>
</ul>
</div>
<div class="io-col out">
<div class="label">EMMA returns</div>
<h3>A calibrated, physics-consistent digital twin.</h3>
<ul>
<li><b>Explicit parameters θ</b><br/><span style="color:var(--ink-3);font-size:14px;">mass, length, damping, thrust and torque coefficients, ...</span></li>
<li><b>Implicit dynamics</b><br/><span style="color:var(--ink-3);font-size:14px;">drag, friction, and other unmeasured physical effects</span></li>
<li><b>Hidden actuation u(t)</b><br/><span style="color:var(--ink-3);font-size:14px;">latent forcing signals such as wheel speed or rotor rate</span></li>
<li><b>Calibration invariants</b><br/><span style="color:var(--ink-3);font-size:14px;">coordinate frame, camera alignment, initial state</span></li>
</ul>
</div>
</div>
</div>
</section>
<hr class="rule" />
<!-- ABSTRACT -->
<section id="abstract">
<div class="wrap narrow">
<div class="kicker">Abstract</div>
<h2>One model, every identifiable parameter.</h2>
<p class="lead">
We introduce EMMA, a physics-informed multimodal framework that recovers all
identifiable dynamical parameters of a system directly from raw video, audio,
and image-based time-series observations. Unlike prior video-only approaches
that struggle with occluded states, hidden actuation inputs, or assumptions
about known initial conditions and coordinate frames, EMMA performs joint
inference of explicit parameters, implicit dynamical components, and
calibration invariants within a unified continuous-time model.
</p>
<p class="lead" style="margin-top: 14px;">
Across 100+ scenarios including five standard dynamical benchmarks (75 Delfys
videos), real-world rover and quadrotor systems with hidden inputs, and
simulation-chart case studies spanning biological and chaotic systems, EMMA
delivers robust multi-parameter recovery and significantly outperforms existing
single-modality and equation-discovery baselines.
</p>
</div>
</section>
<hr class="rule" />
<!-- THE PHYSICS (equations) -->
<section id="physics">
<div class="wrap narrow">
<div class="kicker">The Physics</div>
<h2>EMMA needs your equation of motion.</h2>
<p class="lead">
EMMA is not data-driven equation discovery. The user supplies the parametric
structure of the governing ODE; EMMA solves the <em>inverse</em> problem of
recovering its parameters, along with any latent forcing and invariants, from
multimodal observations.
</p>
<div class="eq-block">
<div class="eq-row">
<div>$$ \frac{d\mathbf{x}(t)}{dt} \;=\; f\!\left(\mathbf{x}(t),\,\mathbf{u}(t);\,\boldsymbol{\theta}\right) $$</div>
<div class="eq-caption">
For a quadrotor, $\mathbf{x}(t)$ is the drone's state (position and velocity),
$\mathbf{u}(t)$ is an external input the camera can't see (wind or rotor thrust),
and $\boldsymbol{\theta}$ is what EMMA recovers (mass, drag, torque).
</div>
</div>
</div>
<p class="lead" style="margin-top: 22px;">
EMMA closes the loop with a physics-informed loss: simulate forward using the
recovered $\boldsymbol{\theta}$, compare to the observed trajectory, and
backpropagate through the ODE solver until the two agree.
</p>
</div>
</section>
<hr class="rule" />
<!-- VIDEO -->
<section id="video">
<div class="wrap narrow">
<div class="kicker">Demo</div>
<h2>Rover, in the lab.</h2>
<p class="lead">
Parameter recovery from a differential-drive rover using synchronized video and
audio. The demo shows the rover scenario only; pendulum, quadrotor, and
simulation-chart results are covered in the paper.
</p>
<a class="yt-card" href="https://youtu.be/Uo79pVlM6Rk" target="_blank" rel="noopener" aria-label="Watch EMMA rover demo on YouTube">
<img src="emma.png" alt="EMMA demo thumbnail"
onerror="this.onerror=null;this.src='https://img.youtube.com/vi/Uo79pVlM6Rk/maxresdefault.jpg';" />
<div class="yt-overlay">
<div class="yt-play" aria-hidden="true"></div>
<div class="yt-badge">EMMA &middot; rover demo</div>
</div>
</a>
</div>
</section>
<hr class="rule" />
<!-- METHOD -->
<section id="method">
<div class="wrap">
<div class="kicker">Method</div>
<h2>Three stages, one continuous-time model.</h2>
<p class="tagline"><span class="em">Sense.</span> <span class="em">Learn.</span> <span class="em">Verify.</span></p>
<p class="lead">
EMMA follows a three-step pipeline from raw data to verified parameters. Video
or chart images are the visual input; audio extends the same pipeline into the
multimodal setting.
</p>
<ol class="stages">
<li>
<div class="step-tag">Step 01 &middot; Sense</div>
<h3>Video &middot; audio &middot; charts</h3>
<p>Raw modalities are converted into time-aligned signals through modality-specific pipelines.</p>
</li>
<li>
<div class="step-tag">Step 02 &middot; Learn</div>
<h3>LTC neural network</h3>
<p>A Liquid Time-Constant network models the system's latent dynamics in continuous time.</p>
</li>
<li>
<div class="step-tag">Step 03 &middot; Verify</div>
<h3>Physics simulation &amp; loss</h3>
<p>A differentiable ODE solver simulates the recovered parameters and checks them against the observations.</p>
</li>
</ol>
<figure class="fig" style="margin-top: 28px;">
<img src="EMMA-arc.png" alt="EMMA architecture: video and audio pipelines feed a shared LTC network and physics-informed digital twin that outputs parameter estimates and a simulated trajectory." />
<figcaption>
<b>Figure 1. EMMA architecture.</b> Multimodal inputs (video, audio, image)
are processed through modality-specific pipelines into unified temporal
representations. An LTC-NN learns latent dynamics <code>h(t)</code> and
predicts physical parameters <code>θ</code>. A differentiable physics solver
validates predictions, enabling end-to-end gradient flow.
</figcaption>
</figure>
</div>
</section>
<hr class="rule" />
<!-- SYSTEMS / EVALUATION -->
<section id="systems">
<div class="wrap">
<div class="kicker">Evaluation</div>
<h2>Across benchmarks, hardware, and simulation charts.</h2>
<p class="lead">
EMMA is evaluated on five canonical video benchmarks, real-world platforms with
hidden forcing, and chart-based simulation studies spanning biological and
chaotic regimes.
</p>
<div class="res-grid">
<div class="res-card">
<div class="res-num">8.8<span>%</span> &plusmn;&nbsp;1.7%</div>
<div class="res-label">mean error</div>
<div class="res-note"><em>Rover (5 params)</em></div>
</div>
<div class="res-card">
<div class="res-num">100<span>+</span> scenarios</div>
<div class="res-label">Delfys benchmark, rover, drone</div>
<div class="res-note"><em>&amp; simulation charts</em></div>
</div>
<div class="res-card">
<div class="res-num">&gt; 10<span>&times;</span> lower</div>
<div class="res-label">implicit RMSE</div>
<div class="res-note"><em>vs PySINDy</em></div>
</div>
<div class="res-card">
<div class="res-num">107<span>&times;</span> smaller</div>
<div class="res-label">compact model</div>
<div class="res-note"><em>vs Delfys baseline</em></div>
</div>
</div>
<p class="lead" style="margin: 28px 0 0;">
Compared against <b>PAIG</b>, <b>NIRPI</b>, and <b>Delfys</b> on the video benchmarks
and <b>PySINDy</b> on the chart-based simulations. Rover and quadrotor under hidden
forcing are reported as first-of-their-kind results. Full tables, ablations, and
per-system comparisons are in the <a href="https://openaccess.thecvf.com/content/CVPR2026/html/Shaikh_EMMA_Extracting_Multiple_physical_parameters_from_Multimodal_Data_CVPR_2026_paper.html" target="_blank" rel="noopener">paper</a>.
</p>
<h3 class="eval-sub">Supported systems</h3>
<div class="sys-grid">
<div class="sys-card">
<div class="cat">Delfys Benchmark</div>
<ul>
<li>Pendulum</li>
<li>Torricelli drainage</li>
<li>Sliding block</li>
<li>LED decay</li>
<li>Free fall</li>
</ul>
</div>
<div class="sys-card">
<div class="cat">Real-world platforms</div>
<ul>
<li>Differential-drive rover</li>
<li>6-DoF quadrotor</li>
<li>Hidden audio forcing</li>
</ul>
</div>
<div class="sys-card">
<div class="cat">Simulation charts</div>
<ul>
<li>Lotka-Volterra</li>
<li>Chaotic Lorenz</li>
<li>F8 Crusader</li>
<li>HIV therapy</li>
<li>AID (Type-1 diabetes)</li>
</ul>
</div>
</div>
</div>
</section>
<hr class="rule" />
<!-- CITATION -->
<section id="citation">
<div class="wrap narrow">
<div class="kicker">Citation</div>
<h2>Cite this work.</h2>
<div class="bibtex-wrap">
<button type="button" class="bibtex-copy" id="bibtex-copy-btn" aria-label="Copy BibTeX to clipboard">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><rect x="9" y="9" width="13" height="13" rx="2" ry="2"/><path d="M5 15H4a2 2 0 0 1-2-2V4a2 2 0 0 1 2-2h9a2 2 0 0 1 2 2v1"/></svg>
<span class="bibtex-copy-label">Copy</span>
</button>
<pre class="bibtex" id="bibtex-source"><span class="k">@InProceedings</span>{Shaikh_2026_CVPR,
<span class="k">author</span> = <span class="v">{Shaikh, Farhat and Banerjee, Ayan and Gupta, Sandeep}</span>,
<span class="k">title</span> = <span class="v">{EMMA: Extracting Multiple physical parameters from Multimodal Data}</span>,
<span class="k">booktitle</span> = <span class="v">{Proceedings of the IEEE/CVF Conference on Computer Vision
and Pattern Recognition (CVPR)}</span>,
<span class="k">month</span> = <span class="v">{June}</span>,
<span class="k">year</span> = <span class="v">{2026}</span>,
<span class="k">pages</span> = <span class="v">{1716-1725}</span>
}</pre>
</div>
</div>
</section>
<!-- FOOTER -->
<footer>
<div class="wrap">
<div class="row">
<div>IMPACT Lab, Arizona State University</div>
<div style="display: flex; gap: 20px;">
<a href="https://github.com/ImpactLabASU/EMMA-CVPR2026" target="_blank" rel="noopener">GitHub</a>
<span>Paper:
<a href="https://openaccess.thecvf.com/content/CVPR2026/html/Shaikh_EMMA_Extracting_Multiple_physical_parameters_from_Multimodal_Data_CVPR_2026_paper.html" target="_blank" rel="noopener">CVF</a>,
<a href="https://arxiv.org/abs/2605.24047" target="_blank" rel="noopener">arXiv</a>
</span>
<a href="#citation">BibTeX</a>
</div>
</div>
<div class="ack">
<strong style="color: var(--ink);">Acknowledgments.</strong> Partially funded by
DARPA AMP (N6600120C4020), DARPA FIRE (P000050426), NSF FDT-Biotech (2436801),
and NIH R21 (1R21HL175632).
</div>
</div>
</footer>
<script>
// Sync all gallery GIFs so they start animating at the same moment.
// Runs once when the #gallery section first becomes visible, then again
// whenever the user scrolls it back into view.
(function () {
var gallery = document.getElementById('gallery');
if (!gallery) return;
var imgs = gallery.querySelectorAll('.gt-pane img');
if (!imgs.length) return;
function resyncAll() {
// Preload first by caching current srcs, then blank+restore in one batch.
var srcs = [];
imgs.forEach(function (img, i) { srcs[i] = img.src; img.src = ''; });
requestAnimationFrame(function () {
imgs.forEach(function (img, i) { img.src = srcs[i]; });
});
}
// Initial sync when gallery comes into view.
if ('IntersectionObserver' in window) {
var io = new IntersectionObserver(function (entries) {
entries.forEach(function (e) { if (e.isIntersecting) resyncAll(); });
}, { threshold: 0.2 });
io.observe(gallery);
} else {
window.addEventListener('load', resyncAll);
}
})();
// Copy BibTeX to clipboard.
(function () {
var btn = document.getElementById('bibtex-copy-btn');
var src = document.getElementById('bibtex-source');
if (!btn || !src) return;
var label = btn.querySelector('.bibtex-copy-label');
var resetTimer = null;
btn.addEventListener('click', function () {
var text = src.innerText || src.textContent || '';
var done = function () {
btn.classList.add('copied');
label.textContent = 'Copied';
clearTimeout(resetTimer);
resetTimer = setTimeout(function () {
btn.classList.remove('copied');
label.textContent = 'Copy';
}, 1600);
};
if (navigator.clipboard && navigator.clipboard.writeText) {
navigator.clipboard.writeText(text).then(done).catch(function () {
fallbackCopy(text, done);
});
} else {
fallbackCopy(text, done);
}
});
function fallbackCopy(text, onDone) {
var ta = document.createElement('textarea');
ta.value = text;
ta.setAttribute('readonly', '');
ta.style.position = 'absolute';
ta.style.left = '-9999px';
document.body.appendChild(ta);
ta.select();
try { document.execCommand('copy'); } catch (e) {}
document.body.removeChild(ta);
onDone();
}
})();
</script>
</body>
</html>