Spaces:
Running
Running
Add static organization landing page
Browse files- index.html +110 -17
index.html
CHANGED
|
@@ -1,19 +1,112 @@
|
|
| 1 |
<!doctype html>
|
| 2 |
-
<html>
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
</html>
|
|
|
|
| 1 |
<!doctype html>
|
| 2 |
+
<html lang="en">
|
| 3 |
+
<head>
|
| 4 |
+
<meta charset="utf-8" />
|
| 5 |
+
<meta name="viewport" content="width=device-width, initial-scale=1" />
|
| 6 |
+
<title>ReasonMatch</title>
|
| 7 |
+
<style>
|
| 8 |
+
:root {
|
| 9 |
+
color-scheme: light dark;
|
| 10 |
+
font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
|
| 11 |
+
line-height: 1.55;
|
| 12 |
+
}
|
| 13 |
+
body {
|
| 14 |
+
margin: 0;
|
| 15 |
+
background: #f7f7fb;
|
| 16 |
+
color: #171717;
|
| 17 |
+
}
|
| 18 |
+
main {
|
| 19 |
+
max-width: 920px;
|
| 20 |
+
margin: 0 auto;
|
| 21 |
+
padding: 48px 24px 64px;
|
| 22 |
+
}
|
| 23 |
+
h1 {
|
| 24 |
+
margin: 0 0 12px;
|
| 25 |
+
font-size: 42px;
|
| 26 |
+
line-height: 1.1;
|
| 27 |
+
}
|
| 28 |
+
h2 {
|
| 29 |
+
margin-top: 36px;
|
| 30 |
+
font-size: 22px;
|
| 31 |
+
}
|
| 32 |
+
p, li {
|
| 33 |
+
font-size: 16px;
|
| 34 |
+
}
|
| 35 |
+
a {
|
| 36 |
+
color: #2454d6;
|
| 37 |
+
}
|
| 38 |
+
.lead {
|
| 39 |
+
max-width: 760px;
|
| 40 |
+
font-size: 18px;
|
| 41 |
+
}
|
| 42 |
+
.links, .resources {
|
| 43 |
+
padding-left: 20px;
|
| 44 |
+
}
|
| 45 |
+
pre {
|
| 46 |
+
overflow-x: auto;
|
| 47 |
+
padding: 16px;
|
| 48 |
+
border-radius: 8px;
|
| 49 |
+
background: #1f2937;
|
| 50 |
+
color: #f9fafb;
|
| 51 |
+
}
|
| 52 |
+
code {
|
| 53 |
+
font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, monospace;
|
| 54 |
+
}
|
| 55 |
+
@media (prefers-color-scheme: dark) {
|
| 56 |
+
body {
|
| 57 |
+
background: #101114;
|
| 58 |
+
color: #f5f5f5;
|
| 59 |
+
}
|
| 60 |
+
a {
|
| 61 |
+
color: #8ab4ff;
|
| 62 |
+
}
|
| 63 |
+
}
|
| 64 |
+
</style>
|
| 65 |
+
</head>
|
| 66 |
+
<body>
|
| 67 |
+
<main>
|
| 68 |
+
<h1>ReasonMatch</h1>
|
| 69 |
+
<p class="lead">
|
| 70 |
+
Official Hugging Face organization for <strong>ReasonMatch-Bench</strong> and the paper
|
| 71 |
+
<strong>Eliciting Complex Spatial Reasoning in MLLMs through Wide-Baseline Matching</strong>.
|
| 72 |
+
</p>
|
| 73 |
+
|
| 74 |
+
<ul class="links">
|
| 75 |
+
<li>Paper: <a href="https://arxiv.org/abs/2606.03577">arXiv:2606.03577</a></li>
|
| 76 |
+
<li>Project page: <a href="https://aim-uofa.github.io/reasonmatch/">https://aim-uofa.github.io/reasonmatch/</a></li>
|
| 77 |
+
<li>Code: <a href="https://github.com/aim-uofa/ReasonMatch">https://github.com/aim-uofa/ReasonMatch</a></li>
|
| 78 |
+
<li>Dataset: <a href="https://www.modelscope.cn/datasets/jxzh2020/ReasonMatchBench">ReasonMatchBench on ModelScope</a></li>
|
| 79 |
+
<li>Pretrained weights: to be announced</li>
|
| 80 |
+
</ul>
|
| 81 |
+
|
| 82 |
+
<h2>About</h2>
|
| 83 |
+
<p>
|
| 84 |
+
ReasonMatch-Bench evaluates visual reasoning over point correspondences. It focuses on
|
| 85 |
+
wide-baseline matching, where models must reason across viewpoint changes, geometric
|
| 86 |
+
transformations, fine-grained visual evidence, and occlusion.
|
| 87 |
+
</p>
|
| 88 |
+
<p>
|
| 89 |
+
The project also includes <strong>Dynamic Correspondence Reinforcement Learning (DCRL)</strong>,
|
| 90 |
+
a verifiable reinforcement learning recipe for wide-baseline matching without explicit
|
| 91 |
+
chain-of-thought supervision.
|
| 92 |
+
</p>
|
| 93 |
+
|
| 94 |
+
<h2>Resources</h2>
|
| 95 |
+
<ul class="resources">
|
| 96 |
+
<li><strong>ReasonMatch-Bench</strong>: in-domain evaluation benchmark for visual matching.</li>
|
| 97 |
+
<li><strong>Out-of-domain evaluation suite</strong>: additional generalization evaluation set.</li>
|
| 98 |
+
<li><strong>Training recipe</strong>: veRL-based DCRL training code for multimodal models.</li>
|
| 99 |
+
<li><strong>Evaluation code</strong>: scripts for ReasonMatch-Bench and out-of-domain evaluation.</li>
|
| 100 |
+
</ul>
|
| 101 |
+
|
| 102 |
+
<h2>Dataset</h2>
|
| 103 |
+
<p>The public evaluation archives are currently hosted on ModelScope:</p>
|
| 104 |
+
<pre><code>dataset_path=/your/absolute/path/to/datasets
|
| 105 |
+
dataset_repo=jxzh2020/ReasonMatchBench
|
| 106 |
+
|
| 107 |
+
modelscope download --repo-type dataset "${dataset_repo}" \
|
| 108 |
+
--include 'reasonmatch_bench.tar.gz' 'ood_dataset.tar.gz' \
|
| 109 |
+
--local_dir "${dataset_path}"</code></pre>
|
| 110 |
+
</main>
|
| 111 |
+
</body>
|
| 112 |
</html>
|