Spaces:
Running
Running
Commit
·
f737afb
1
Parent(s):
4796e93
Add RTC experiments visualization website with Git LFS
Browse files- Set up Git LFS for PNG files
- Added interactive experiment viewer with tab navigation
- Showcases 57 experiments across PI0.5 and SmolVLA models
- 285 total visualizations with different flow matching steps and sigma_d values
- Responsive grid layout with fullscreen modal support
This view is limited to 50 files because it contains too many changes.
See raw diff
- .gitattributes +1 -0
- index.html +372 -374
- static/experiments/experiment_log.txt +378 -0
- static/experiments/pi05_steps_10_sigma_0.1/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.1/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.1/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.1/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.1/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.2/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.2/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.2/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.2/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.2/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.5/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.5/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.5/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.5/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_10_sigma_0.5/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.1/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.1/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.1/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.1/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.1/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.2/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.2/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.2/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.2/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.2/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.5/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.5/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.5/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.5/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.5/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.8/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.8/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.8/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.8/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_0.8/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.0/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.0/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.0/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.0/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.0/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.2/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.2/denoising_vt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.2/denoising_x1t_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.2/denoising_xt_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.2/final_actions_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.5/denoising_correction_comparison.png +3 -0
- static/experiments/pi05_steps_2_sigma_1.5/denoising_vt_comparison.png +3 -0
.gitattributes
CHANGED
|
@@ -46,3 +46,4 @@ static/videos/shiba.mp4 filter=lfs diff=lfs merge=lfs -text
|
|
| 46 |
static/videos/steve.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 47 |
static/videos/teaser.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 48 |
static/videos/toby.mp4 filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 46 |
static/videos/steve.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 47 |
static/videos/teaser.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 48 |
static/videos/toby.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 49 |
+
*.png filter=lfs diff=lfs merge=lfs -text
|
index.html
CHANGED
|
@@ -2,29 +2,170 @@
|
|
| 2 |
<html>
|
| 3 |
<head>
|
| 4 |
<meta charset="utf-8">
|
| 5 |
-
<meta name="description"
|
| 6 |
-
|
| 7 |
-
<meta name="keywords" content="Nerfies, D-NeRF, NeRF">
|
| 8 |
<meta name="viewport" content="width=device-width, initial-scale=1">
|
| 9 |
-
<title>
|
| 10 |
-
|
| 11 |
-
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
|
| 12 |
-
rel="stylesheet">
|
| 13 |
|
|
|
|
| 14 |
<link rel="stylesheet" href="./static/css/bulma.min.css">
|
| 15 |
-
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
|
| 16 |
-
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
|
| 17 |
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
|
| 18 |
-
<link rel="stylesheet"
|
| 19 |
-
href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
|
| 20 |
<link rel="stylesheet" href="./static/css/index.css">
|
| 21 |
<link rel="icon" href="./static/images/favicon.svg">
|
| 22 |
|
| 23 |
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
|
| 24 |
<script defer src="./static/js/fontawesome.all.min.js"></script>
|
| 25 |
-
|
| 26 |
-
<
|
| 27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
</head>
|
| 29 |
<body>
|
| 30 |
|
|
@@ -33,86 +174,10 @@
|
|
| 33 |
<div class="container is-max-desktop">
|
| 34 |
<div class="columns is-centered">
|
| 35 |
<div class="column has-text-centered">
|
| 36 |
-
<h1 class="title is-1 publication-title">
|
| 37 |
-
<
|
| 38 |
-
<span class="author-block">
|
| 39 |
-
<a href="https://keunhong.com" target="_blank">Keunhong Park</a><sup>1</sup>,</span>
|
| 40 |
-
<span class="author-block">
|
| 41 |
-
<a href="https://utkarshsinha.com" target="_blank">Utkarsh Sinha</a><sup>2</sup>,</span>
|
| 42 |
-
<span class="author-block">
|
| 43 |
-
<a href="https://jonbarron.info" target="_blank">Jonathan T. Barron</a><sup>2</sup>,
|
| 44 |
-
</span>
|
| 45 |
-
<span class="author-block">
|
| 46 |
-
<a href="http://sofienbouaziz.com" target="_blank">Sofien Bouaziz</a><sup>2</sup>,
|
| 47 |
-
</span>
|
| 48 |
-
<span class="author-block">
|
| 49 |
-
<a href="https://www.danbgoldman.com" target="_blank">Dan B Goldman</a><sup>2</sup>,
|
| 50 |
-
</span>
|
| 51 |
-
<span class="author-block">
|
| 52 |
-
<a href="https://homes.cs.washington.edu/~seitz/" target="_blank">Steven M. Seitz</a><sup>1,2</sup>,
|
| 53 |
-
</span>
|
| 54 |
-
<span class="author-block">
|
| 55 |
-
<a href="http://www.ricardomartinbrualla.com" target="_blank">Ricardo Martin-Brualla</a><sup>2</sup>
|
| 56 |
-
</span>
|
| 57 |
-
</div>
|
| 58 |
-
|
| 59 |
<div class="is-size-5 publication-authors">
|
| 60 |
-
<span class="author-block"
|
| 61 |
-
<span class="author-block"><sup>2</sup>Google Research</span>
|
| 62 |
-
</div>
|
| 63 |
-
|
| 64 |
-
<div class="column has-text-centered">
|
| 65 |
-
<div class="publication-links">
|
| 66 |
-
<!-- PDF Link. -->
|
| 67 |
-
<span class="link-block">
|
| 68 |
-
<a href="https://arxiv.org/pdf/2011.12948" target="_blank"
|
| 69 |
-
class="external-link button is-normal is-rounded is-dark">
|
| 70 |
-
<span class="icon">
|
| 71 |
-
<i class="fas fa-file-pdf"></i>
|
| 72 |
-
</span>
|
| 73 |
-
<span>Paper</span>
|
| 74 |
-
</a>
|
| 75 |
-
</span>
|
| 76 |
-
<span class="link-block">
|
| 77 |
-
<a href="https://arxiv.org/abs/2011.12948" target="_blank"
|
| 78 |
-
class="external-link button is-normal is-rounded is-dark">
|
| 79 |
-
<span class="icon">
|
| 80 |
-
<i class="ai ai-arxiv"></i>
|
| 81 |
-
</span>
|
| 82 |
-
<span>arXiv</span>
|
| 83 |
-
</a>
|
| 84 |
-
</span>
|
| 85 |
-
<!-- Video Link. -->
|
| 86 |
-
<span class="link-block">
|
| 87 |
-
<a href="https://www.youtube.com/watch?v=MrKrnHhk8IA" target="_blank"
|
| 88 |
-
class="external-link button is-normal is-rounded is-dark">
|
| 89 |
-
<span class="icon">
|
| 90 |
-
<i class="fab fa-youtube"></i>
|
| 91 |
-
</span>
|
| 92 |
-
<span>Video</span>
|
| 93 |
-
</a>
|
| 94 |
-
</span>
|
| 95 |
-
<!-- Code Link. -->
|
| 96 |
-
<span class="link-block">
|
| 97 |
-
<a href="https://github.com/google/nerfies" target="_blank"
|
| 98 |
-
class="external-link button is-normal is-rounded is-dark">
|
| 99 |
-
<span class="icon">
|
| 100 |
-
<i class="fab fa-github"></i>
|
| 101 |
-
</span>
|
| 102 |
-
<span>Code</span>
|
| 103 |
-
</a>
|
| 104 |
-
</span>
|
| 105 |
-
<!-- Dataset Link. -->
|
| 106 |
-
<span class="link-block">
|
| 107 |
-
<a href="https://github.com/google/nerfies/releases/tag/0.1" target="_blank"
|
| 108 |
-
class="external-link button is-normal is-rounded is-dark">
|
| 109 |
-
<span class="icon">
|
| 110 |
-
<i class="far fa-images"></i>
|
| 111 |
-
</span>
|
| 112 |
-
<span>Data</span>
|
| 113 |
-
</a>
|
| 114 |
-
</div>
|
| 115 |
-
|
| 116 |
</div>
|
| 117 |
</div>
|
| 118 |
</div>
|
|
@@ -120,316 +185,249 @@
|
|
| 120 |
</div>
|
| 121 |
</section>
|
| 122 |
|
| 123 |
-
<section class="
|
| 124 |
<div class="container is-max-desktop">
|
| 125 |
-
<div class="
|
| 126 |
-
<
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 135 |
</div>
|
| 136 |
-
</div>
|
| 137 |
-
</section>
|
| 138 |
|
| 139 |
-
|
| 140 |
-
<
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
<div id="results-carousel" class="carousel results-carousel">
|
| 144 |
-
<div class="item item-steve">
|
| 145 |
-
<video poster="" id="steve" autoplay controls muted loop playsinline height="100%">
|
| 146 |
-
<source src="./static/videos/steve.mp4"
|
| 147 |
-
type="video/mp4">
|
| 148 |
-
</video>
|
| 149 |
-
</div>
|
| 150 |
-
<div class="item item-chair-tp">
|
| 151 |
-
<video poster="" id="chair-tp" autoplay controls muted loop playsinline height="100%">
|
| 152 |
-
<source src="./static/videos/chair-tp.mp4"
|
| 153 |
-
type="video/mp4">
|
| 154 |
-
</video>
|
| 155 |
-
</div>
|
| 156 |
-
<div class="item item-shiba">
|
| 157 |
-
<video poster="" id="shiba" autoplay controls muted loop playsinline height="100%">
|
| 158 |
-
<source src="./static/videos/shiba.mp4"
|
| 159 |
-
type="video/mp4">
|
| 160 |
-
</video>
|
| 161 |
-
</div>
|
| 162 |
-
<div class="item item-fullbody">
|
| 163 |
-
<video poster="" id="fullbody" autoplay controls muted loop playsinline height="100%">
|
| 164 |
-
<source src="./static/videos/fullbody.mp4"
|
| 165 |
-
type="video/mp4">
|
| 166 |
-
</video>
|
| 167 |
-
</div>
|
| 168 |
-
<div class="item item-blueshirt">
|
| 169 |
-
<video poster="" id="blueshirt" autoplay controls muted loop playsinline height="100%">
|
| 170 |
-
<source src="./static/videos/blueshirt.mp4"
|
| 171 |
-
type="video/mp4">
|
| 172 |
-
</video>
|
| 173 |
-
</div>
|
| 174 |
-
<div class="item item-mask">
|
| 175 |
-
<video poster="" id="mask" autoplay controls muted loop playsinline height="100%">
|
| 176 |
-
<source src="./static/videos/mask.mp4"
|
| 177 |
-
type="video/mp4">
|
| 178 |
-
</video>
|
| 179 |
-
</div>
|
| 180 |
-
<div class="item item-coffee">
|
| 181 |
-
<video poster="" id="coffee" autoplay controls muted loop playsinline height="100%">
|
| 182 |
-
<source src="./static/videos/coffee.mp4"
|
| 183 |
-
type="video/mp4">
|
| 184 |
-
</video>
|
| 185 |
-
</div>
|
| 186 |
-
<div class="item item-toby">
|
| 187 |
-
<video poster="" id="toby" autoplay controls muted loop playsinline height="100%">
|
| 188 |
-
<source src="./static/videos/toby2.mp4"
|
| 189 |
-
type="video/mp4">
|
| 190 |
-
</video>
|
| 191 |
-
</div>
|
| 192 |
-
</div>
|
| 193 |
</div>
|
| 194 |
-
</div>
|
| 195 |
-
</section>
|
| 196 |
|
|
|
|
|
|
|
|
|
|
| 197 |
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
|
| 201 |
-
|
| 202 |
-
<div class="column is-four-fifths">
|
| 203 |
-
<h2 class="title is-3">Abstract</h2>
|
| 204 |
-
<div class="content has-text-justified">
|
| 205 |
-
<p>
|
| 206 |
-
We present the first method capable of photorealistically reconstructing a non-rigidly
|
| 207 |
-
deforming scene using photos/videos captured casually from mobile phones.
|
| 208 |
-
</p>
|
| 209 |
-
<p>
|
| 210 |
-
Our approach augments neural radiance fields
|
| 211 |
-
(NeRF) by optimizing an
|
| 212 |
-
additional continuous volumetric deformation field that warps each observed point into a
|
| 213 |
-
canonical 5D NeRF.
|
| 214 |
-
We observe that these NeRF-like deformation fields are prone to local minima, and
|
| 215 |
-
propose a coarse-to-fine optimization method for coordinate-based models that allows for
|
| 216 |
-
more robust optimization.
|
| 217 |
-
By adapting principles from geometry processing and physical simulation to NeRF-like
|
| 218 |
-
models, we propose an elastic regularization of the deformation field that further
|
| 219 |
-
improves robustness.
|
| 220 |
-
</p>
|
| 221 |
-
<p>
|
| 222 |
-
We show that <span class="dnerf">Nerfies</span> can turn casually captured selfie
|
| 223 |
-
photos/videos into deformable NeRF
|
| 224 |
-
models that allow for photorealistic renderings of the subject from arbitrary
|
| 225 |
-
viewpoints, which we dub <i>"nerfies"</i>. We evaluate our method by collecting data
|
| 226 |
-
using a
|
| 227 |
-
rig with two mobile phones that take time-synchronized photos, yielding train/validation
|
| 228 |
-
images of the same pose at different viewpoints. We show that our method faithfully
|
| 229 |
-
reconstructs non-rigidly deforming scenes and reproduces unseen views with high
|
| 230 |
-
fidelity.
|
| 231 |
-
</p>
|
| 232 |
-
</div>
|
| 233 |
</div>
|
| 234 |
-
|
| 235 |
-
|
| 236 |
-
|
| 237 |
-
|
| 238 |
-
|
| 239 |
-
<div class="column is-four-fifths">
|
| 240 |
-
<h2 class="title is-3">Video</h2>
|
| 241 |
-
<div class="publication-video">
|
| 242 |
-
<iframe src="https://www.youtube.com/embed/MrKrnHhk8IA?rel=0&showinfo=0"
|
| 243 |
-
frameborder="0" allow="autoplay; encrypted-media" allowfullscreen></iframe>
|
| 244 |
-
</div>
|
| 245 |
</div>
|
| 246 |
-
</div>
|
| 247 |
-
<!--/ Paper video. -->
|
| 248 |
-
</div>
|
| 249 |
-
</section>
|
| 250 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 251 |
|
| 252 |
-
|
| 253 |
-
|
|
|
|
| 254 |
|
| 255 |
-
|
| 256 |
-
|
| 257 |
-
|
| 258 |
-
|
| 259 |
-
<div class="content">
|
| 260 |
-
<h2 class="title is-3">Visual Effects</h2>
|
| 261 |
-
<p>
|
| 262 |
-
Using <i>nerfies</i> you can create fun visual effects. This Dolly zoom effect
|
| 263 |
-
would be impossible without nerfies since it would require going through a wall.
|
| 264 |
-
</p>
|
| 265 |
-
<video id="dollyzoom" autoplay controls muted loop playsinline height="100%">
|
| 266 |
-
<source src="./static/videos/dollyzoom-stacked.mp4"
|
| 267 |
-
type="video/mp4">
|
| 268 |
-
</video>
|
| 269 |
-
</div>
|
| 270 |
</div>
|
| 271 |
-
<!--/ Visual Effects. -->
|
| 272 |
-
|
| 273 |
-
<!-- Matting. -->
|
| 274 |
-
<div class="column">
|
| 275 |
-
<h2 class="title is-3">Matting</h2>
|
| 276 |
-
<div class="columns is-centered">
|
| 277 |
-
<div class="column content">
|
| 278 |
-
<p>
|
| 279 |
-
As a byproduct of our method, we can also solve the matting problem by ignoring
|
| 280 |
-
samples that fall outside of a bounding box during rendering.
|
| 281 |
-
</p>
|
| 282 |
-
<video id="matting-video" controls playsinline height="100%">
|
| 283 |
-
<source src="./static/videos/matting.mp4"
|
| 284 |
-
type="video/mp4">
|
| 285 |
-
</video>
|
| 286 |
-
</div>
|
| 287 |
|
| 288 |
-
|
|
|
|
|
|
|
|
|
|
| 289 |
</div>
|
| 290 |
-
</div>
|
| 291 |
-
<!--/ Matting. -->
|
| 292 |
-
|
| 293 |
-
<!-- Animation. -->
|
| 294 |
-
<div class="columns is-centered">
|
| 295 |
-
<div class="column is-full-width">
|
| 296 |
-
<h2 class="title is-3">Animation</h2>
|
| 297 |
-
|
| 298 |
-
<!-- Interpolating. -->
|
| 299 |
-
<h3 class="title is-4">Interpolating states</h3>
|
| 300 |
-
<div class="content has-text-justified">
|
| 301 |
-
<p>
|
| 302 |
-
We can also animate the scene by interpolating the deformation latent codes of two input
|
| 303 |
-
frames. Use the slider here to linearly interpolate between the left frame and the right
|
| 304 |
-
frame.
|
| 305 |
-
</p>
|
| 306 |
-
</div>
|
| 307 |
-
<div class="columns is-vcentered interpolation-panel">
|
| 308 |
-
<div class="column is-3 has-text-centered">
|
| 309 |
-
<img src="./static/images/interpolate_start.jpg"
|
| 310 |
-
class="interpolation-image"
|
| 311 |
-
alt="Interpolate start reference image."/>
|
| 312 |
-
<p>Start Frame</p>
|
| 313 |
-
</div>
|
| 314 |
-
<div class="column interpolation-video-column">
|
| 315 |
-
<div id="interpolation-image-wrapper">
|
| 316 |
-
Loading...
|
| 317 |
-
</div>
|
| 318 |
-
<input class="slider is-fullwidth is-large is-info"
|
| 319 |
-
id="interpolation-slider"
|
| 320 |
-
step="1" min="0" max="100" value="0" type="range">
|
| 321 |
-
</div>
|
| 322 |
-
<div class="column is-3 has-text-centered">
|
| 323 |
-
<img src="./static/images/interpolate_end.jpg"
|
| 324 |
-
class="interpolation-image"
|
| 325 |
-
alt="Interpolation end reference image."/>
|
| 326 |
-
<p class="is-bold">End Frame</p>
|
| 327 |
-
</div>
|
| 328 |
-
</div>
|
| 329 |
-
<br/>
|
| 330 |
-
<!--/ Interpolating. -->
|
| 331 |
-
|
| 332 |
-
<!-- Re-rendering. -->
|
| 333 |
-
<h3 class="title is-4">Re-rendering the input video</h3>
|
| 334 |
-
<div class="content has-text-justified">
|
| 335 |
-
<p>
|
| 336 |
-
Using <span class="dnerf">Nerfies</span>, you can re-render a video from a novel
|
| 337 |
-
viewpoint such as a stabilized camera by playing back the training deformations.
|
| 338 |
-
</p>
|
| 339 |
-
</div>
|
| 340 |
-
<div class="content has-text-centered">
|
| 341 |
-
<video id="replay-video"
|
| 342 |
-
controls
|
| 343 |
-
muted
|
| 344 |
-
preload
|
| 345 |
-
playsinline
|
| 346 |
-
width="75%">
|
| 347 |
-
<source src="./static/videos/replay.mp4"
|
| 348 |
-
type="video/mp4">
|
| 349 |
-
</video>
|
| 350 |
-
</div>
|
| 351 |
-
<!--/ Re-rendering. -->
|
| 352 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 353 |
</div>
|
| 354 |
-
</div>
|
| 355 |
-
<!--/ Animation. -->
|
| 356 |
-
|
| 357 |
-
|
| 358 |
-
<!-- Concurrent Work. -->
|
| 359 |
-
<div class="columns is-centered">
|
| 360 |
-
<div class="column is-full-width">
|
| 361 |
-
<h2 class="title is-3">Related Links</h2>
|
| 362 |
-
|
| 363 |
-
<div class="content has-text-justified">
|
| 364 |
-
<p>
|
| 365 |
-
There's a lot of excellent work that was introduced around the same time as ours.
|
| 366 |
-
</p>
|
| 367 |
-
<p>
|
| 368 |
-
<a href="https://arxiv.org/abs/2104.09125" target="_blank">Progressive Encoding for Neural Optimization</a> introduces an idea similar to our windowed position encoding for coarse-to-fine optimization.
|
| 369 |
-
</p>
|
| 370 |
-
<p>
|
| 371 |
-
<a href="https://www.albertpumarola.com/research/D-NeRF/index.html" target="_blank">D-NeRF</a> and <a href="https://gvv.mpi-inf.mpg.de/projects/nonrigid_nerf/" target="_blank">NR-NeRF</a>
|
| 372 |
-
both use deformation fields to model non-rigid scenes.
|
| 373 |
-
</p>
|
| 374 |
-
<p>
|
| 375 |
-
Some works model videos with a NeRF by directly modulating the density, such as <a href="https://video-nerf.github.io/" target="_blank">Video-NeRF</a>, <a href="https://www.cs.cornell.edu/~zl548/NSFF/" target="_blank">NSFF</a>, and <a href="https://neural-3d-video.github.io/" target="_blank">DyNeRF</a>
|
| 376 |
-
</p>
|
| 377 |
-
<p>
|
| 378 |
-
There are probably many more by the time you are reading this. Check out <a href="https://dellaert.github.io/NeRF/" target="_blank">Frank Dellart's survey on recent NeRF papers</a>, and <a href="https://github.com/yenchenlin/awesome-NeRF" target="_blank">Yen-Chen Lin's curated list of NeRF papers</a>.
|
| 379 |
-
</p>
|
| 380 |
-
</div>
|
| 381 |
-
</div>
|
| 382 |
-
</div>
|
| 383 |
-
<!--/ Concurrent Work. -->
|
| 384 |
|
| 385 |
-
|
| 386 |
-
|
|
|
|
|
|
|
|
|
|
| 387 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 388 |
|
| 389 |
-
|
| 390 |
-
|
| 391 |
-
|
| 392 |
-
|
| 393 |
-
|
| 394 |
-
|
| 395 |
-
journal = {ICCV},
|
| 396 |
-
year = {2021},
|
| 397 |
-
}</code></pre>
|
| 398 |
</div>
|
| 399 |
</section>
|
| 400 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 401 |
|
| 402 |
<footer class="footer">
|
| 403 |
<div class="container">
|
| 404 |
<div class="content has-text-centered">
|
| 405 |
-
<
|
| 406 |
-
|
| 407 |
-
|
| 408 |
-
</a>
|
| 409 |
-
<a class="icon-link" href="https://github.com/keunhong" target="_blank" class="external-link" disabled>
|
| 410 |
-
<i class="fab fa-github"></i>
|
| 411 |
-
</a>
|
| 412 |
-
</div>
|
| 413 |
-
<div class="columns is-centered">
|
| 414 |
-
<div class="column is-8">
|
| 415 |
-
<div class="content">
|
| 416 |
-
<p>
|
| 417 |
-
This website is licensed under a <a rel="license" target="_blank"
|
| 418 |
-
href="http://creativecommons.org/licenses/by-sa/4.0/">Creative
|
| 419 |
-
Commons Attribution-ShareAlike 4.0 International License</a>.
|
| 420 |
-
</p>
|
| 421 |
-
<p>
|
| 422 |
-
This means you are free to borrow the <a target="_blank"
|
| 423 |
-
href="https://github.com/nerfies/nerfies.github.io">source code</a> of this website,
|
| 424 |
-
we just ask that you link back to this page in the footer.
|
| 425 |
-
Please remember to remove the analytics code included in the header of the website which
|
| 426 |
-
you do not want on your website.
|
| 427 |
-
</p>
|
| 428 |
-
</div>
|
| 429 |
-
</div>
|
| 430 |
</div>
|
| 431 |
</div>
|
| 432 |
</footer>
|
| 433 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 434 |
</body>
|
| 435 |
</html>
|
|
|
|
| 2 |
<html>
|
| 3 |
<head>
|
| 4 |
<meta charset="utf-8">
|
| 5 |
+
<meta name="description" content="Real-Time Chunking (RTC) Experiments - Flow Matching Steps and Sigma_d Parameter Analysis">
|
| 6 |
+
<meta name="keywords" content="RTC, Real-Time Chunking, LeRobot, PI0.5, SmolVLA">
|
|
|
|
| 7 |
<meta name="viewport" content="width=device-width, initial-scale=1">
|
| 8 |
+
<title>RTC Experiments: Flow Matching Steps & Sigma Analysis</title>
|
|
|
|
|
|
|
|
|
|
| 9 |
|
| 10 |
+
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
|
| 11 |
<link rel="stylesheet" href="./static/css/bulma.min.css">
|
|
|
|
|
|
|
| 12 |
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
|
| 13 |
+
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
|
|
|
|
| 14 |
<link rel="stylesheet" href="./static/css/index.css">
|
| 15 |
<link rel="icon" href="./static/images/favicon.svg">
|
| 16 |
|
| 17 |
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
|
| 18 |
<script defer src="./static/js/fontawesome.all.min.js"></script>
|
| 19 |
+
|
| 20 |
+
<style>
|
| 21 |
+
.experiment-grid {
|
| 22 |
+
display: grid;
|
| 23 |
+
grid-template-columns: repeat(auto-fit, minmax(300px, 1fr));
|
| 24 |
+
gap: 20px;
|
| 25 |
+
margin-top: 20px;
|
| 26 |
+
}
|
| 27 |
+
|
| 28 |
+
.experiment-card {
|
| 29 |
+
border: 1px solid #ddd;
|
| 30 |
+
border-radius: 8px;
|
| 31 |
+
padding: 15px;
|
| 32 |
+
background: white;
|
| 33 |
+
box-shadow: 0 2px 4px rgba(0,0,0,0.1);
|
| 34 |
+
transition: transform 0.2s;
|
| 35 |
+
}
|
| 36 |
+
|
| 37 |
+
.experiment-card:hover {
|
| 38 |
+
transform: translateY(-5px);
|
| 39 |
+
box-shadow: 0 4px 8px rgba(0,0,0,0.2);
|
| 40 |
+
}
|
| 41 |
+
|
| 42 |
+
.experiment-card h3 {
|
| 43 |
+
margin-bottom: 10px;
|
| 44 |
+
color: #363636;
|
| 45 |
+
font-size: 1.1rem;
|
| 46 |
+
font-weight: 600;
|
| 47 |
+
}
|
| 48 |
+
|
| 49 |
+
.experiment-card img {
|
| 50 |
+
width: 100%;
|
| 51 |
+
height: auto;
|
| 52 |
+
border-radius: 4px;
|
| 53 |
+
margin-bottom: 10px;
|
| 54 |
+
cursor: pointer;
|
| 55 |
+
}
|
| 56 |
+
|
| 57 |
+
.visualization-tabs {
|
| 58 |
+
display: flex;
|
| 59 |
+
flex-wrap: wrap;
|
| 60 |
+
gap: 5px;
|
| 61 |
+
margin-bottom: 10px;
|
| 62 |
+
}
|
| 63 |
+
|
| 64 |
+
.visualization-tab {
|
| 65 |
+
padding: 5px 10px;
|
| 66 |
+
background: #f5f5f5;
|
| 67 |
+
border: 1px solid #ddd;
|
| 68 |
+
border-radius: 4px;
|
| 69 |
+
cursor: pointer;
|
| 70 |
+
font-size: 0.85rem;
|
| 71 |
+
transition: all 0.2s;
|
| 72 |
+
}
|
| 73 |
+
|
| 74 |
+
.visualization-tab:hover {
|
| 75 |
+
background: #e8e8e8;
|
| 76 |
+
}
|
| 77 |
+
|
| 78 |
+
.visualization-tab.active {
|
| 79 |
+
background: #3273dc;
|
| 80 |
+
color: white;
|
| 81 |
+
border-color: #3273dc;
|
| 82 |
+
}
|
| 83 |
+
|
| 84 |
+
.section-header {
|
| 85 |
+
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
|
| 86 |
+
color: white;
|
| 87 |
+
padding: 20px;
|
| 88 |
+
border-radius: 8px;
|
| 89 |
+
margin-bottom: 30px;
|
| 90 |
+
}
|
| 91 |
+
|
| 92 |
+
.model-section {
|
| 93 |
+
margin-bottom: 60px;
|
| 94 |
+
}
|
| 95 |
+
|
| 96 |
+
.steps-group {
|
| 97 |
+
margin-bottom: 40px;
|
| 98 |
+
padding: 20px;
|
| 99 |
+
background: #f9f9f9;
|
| 100 |
+
border-radius: 8px;
|
| 101 |
+
}
|
| 102 |
+
|
| 103 |
+
.steps-group h2 {
|
| 104 |
+
color: #363636;
|
| 105 |
+
margin-bottom: 20px;
|
| 106 |
+
padding-bottom: 10px;
|
| 107 |
+
border-bottom: 2px solid #3273dc;
|
| 108 |
+
}
|
| 109 |
+
|
| 110 |
+
.image-modal {
|
| 111 |
+
display: none;
|
| 112 |
+
position: fixed;
|
| 113 |
+
z-index: 1000;
|
| 114 |
+
left: 0;
|
| 115 |
+
top: 0;
|
| 116 |
+
width: 100%;
|
| 117 |
+
height: 100%;
|
| 118 |
+
background-color: rgba(0,0,0,0.9);
|
| 119 |
+
}
|
| 120 |
+
|
| 121 |
+
.image-modal img {
|
| 122 |
+
margin: auto;
|
| 123 |
+
display: block;
|
| 124 |
+
max-width: 90%;
|
| 125 |
+
max-height: 90%;
|
| 126 |
+
position: absolute;
|
| 127 |
+
top: 50%;
|
| 128 |
+
left: 50%;
|
| 129 |
+
transform: translate(-50%, -50%);
|
| 130 |
+
}
|
| 131 |
+
|
| 132 |
+
.close-modal {
|
| 133 |
+
position: absolute;
|
| 134 |
+
top: 20px;
|
| 135 |
+
right: 35px;
|
| 136 |
+
color: #f1f1f1;
|
| 137 |
+
font-size: 40px;
|
| 138 |
+
font-weight: bold;
|
| 139 |
+
cursor: pointer;
|
| 140 |
+
}
|
| 141 |
+
|
| 142 |
+
.nav-tabs {
|
| 143 |
+
display: flex;
|
| 144 |
+
gap: 10px;
|
| 145 |
+
margin-bottom: 20px;
|
| 146 |
+
flex-wrap: wrap;
|
| 147 |
+
}
|
| 148 |
+
|
| 149 |
+
.nav-tab {
|
| 150 |
+
padding: 10px 20px;
|
| 151 |
+
background: #f5f5f5;
|
| 152 |
+
border: 2px solid #ddd;
|
| 153 |
+
border-radius: 6px;
|
| 154 |
+
cursor: pointer;
|
| 155 |
+
font-weight: 600;
|
| 156 |
+
transition: all 0.3s;
|
| 157 |
+
}
|
| 158 |
+
|
| 159 |
+
.nav-tab:hover {
|
| 160 |
+
background: #e8e8e8;
|
| 161 |
+
}
|
| 162 |
+
|
| 163 |
+
.nav-tab.active {
|
| 164 |
+
background: #3273dc;
|
| 165 |
+
color: white;
|
| 166 |
+
border-color: #3273dc;
|
| 167 |
+
}
|
| 168 |
+
</style>
|
| 169 |
</head>
|
| 170 |
<body>
|
| 171 |
|
|
|
|
| 174 |
<div class="container is-max-desktop">
|
| 175 |
<div class="columns is-centered">
|
| 176 |
<div class="column has-text-centered">
|
| 177 |
+
<h1 class="title is-1 publication-title">Real-Time Chunking Experiments</h1>
|
| 178 |
+
<h2 class="subtitle is-3">Flow Matching Steps & Sigma_d Parameter Analysis</h2>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 179 |
<div class="is-size-5 publication-authors">
|
| 180 |
+
<span class="author-block">LeRobot RTC Integration Tests</span>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 181 |
</div>
|
| 182 |
</div>
|
| 183 |
</div>
|
|
|
|
| 185 |
</div>
|
| 186 |
</section>
|
| 187 |
|
| 188 |
+
<section class="section">
|
| 189 |
<div class="container is-max-desktop">
|
| 190 |
+
<div class="section-header">
|
| 191 |
+
<h2 class="title is-3" style="color: white; margin-bottom: 10px;">Experiment Overview</h2>
|
| 192 |
+
<div class="content has-text-justified" style="color: white;">
|
| 193 |
+
<p>
|
| 194 |
+
This page presents comprehensive results from Real-Time Chunking (RTC) experiments across different models,
|
| 195 |
+
flow matching steps, and sigma_d values. The experiments demonstrate how these parameters affect the
|
| 196 |
+
denoising process and final action predictions.
|
| 197 |
+
</p>
|
| 198 |
+
<p>
|
| 199 |
+
<strong>Models tested:</strong> PI0.5, SmolVLA<br>
|
| 200 |
+
<strong>Flow matching steps:</strong> 2, 5, 10, 20, 50, 100 (varies by model)<br>
|
| 201 |
+
<strong>Sigma_d values:</strong> 0.1, 0.2, 0.5, 0.8, 1.0, 1.2, 1.5
|
| 202 |
+
</p>
|
| 203 |
+
</div>
|
| 204 |
</div>
|
|
|
|
|
|
|
| 205 |
|
| 206 |
+
<!-- Navigation Tabs -->
|
| 207 |
+
<div class="nav-tabs">
|
| 208 |
+
<div class="nav-tab active" onclick="showModel('pi05')">PI0.5 Model</div>
|
| 209 |
+
<div class="nav-tab" onclick="showModel('smolvla')">SmolVLA Model</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 210 |
</div>
|
|
|
|
|
|
|
| 211 |
|
| 212 |
+
<!-- PI0.5 Model Section -->
|
| 213 |
+
<div id="pi05-section" class="model-section">
|
| 214 |
+
<h1 class="title is-2">PI0.5 Model Results</h1>
|
| 215 |
|
| 216 |
+
<!-- Steps 2 -->
|
| 217 |
+
<div class="steps-group">
|
| 218 |
+
<h2 class="title is-3">Flow Matching Steps: 2</h2>
|
| 219 |
+
<div class="experiment-grid" id="pi05-steps-2"></div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 220 |
</div>
|
| 221 |
+
|
| 222 |
+
<!-- Steps 5 -->
|
| 223 |
+
<div class="steps-group">
|
| 224 |
+
<h2 class="title is-3">Flow Matching Steps: 5</h2>
|
| 225 |
+
<div class="experiment-grid" id="pi05-steps-5"></div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 226 |
</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
| 227 |
|
| 228 |
+
<!-- Steps 10 -->
|
| 229 |
+
<div class="steps-group">
|
| 230 |
+
<h2 class="title is-3">Flow Matching Steps: 10</h2>
|
| 231 |
+
<div class="experiment-grid" id="pi05-steps-10"></div>
|
| 232 |
+
</div>
|
| 233 |
+
</div>
|
| 234 |
|
| 235 |
+
<!-- SmolVLA Model Section -->
|
| 236 |
+
<div id="smolvla-section" class="model-section" style="display: none;">
|
| 237 |
+
<h1 class="title is-2">SmolVLA Model Results</h1>
|
| 238 |
|
| 239 |
+
<!-- Steps 2 -->
|
| 240 |
+
<div class="steps-group">
|
| 241 |
+
<h2 class="title is-3">Flow Matching Steps: 2</h2>
|
| 242 |
+
<div class="experiment-grid" id="smolvla-steps-2"></div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 243 |
</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 244 |
|
| 245 |
+
<!-- Steps 5 -->
|
| 246 |
+
<div class="steps-group">
|
| 247 |
+
<h2 class="title is-3">Flow Matching Steps: 5</h2>
|
| 248 |
+
<div class="experiment-grid" id="smolvla-steps-5"></div>
|
| 249 |
</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 250 |
|
| 251 |
+
<!-- Steps 10 -->
|
| 252 |
+
<div class="steps-group">
|
| 253 |
+
<h2 class="title is-3">Flow Matching Steps: 10</h2>
|
| 254 |
+
<div class="experiment-grid" id="smolvla-steps-10"></div>
|
| 255 |
</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 256 |
|
| 257 |
+
<!-- Steps 20 -->
|
| 258 |
+
<div class="steps-group">
|
| 259 |
+
<h2 class="title is-3">Flow Matching Steps: 20</h2>
|
| 260 |
+
<div class="experiment-grid" id="smolvla-steps-20"></div>
|
| 261 |
+
</div>
|
| 262 |
|
| 263 |
+
<!-- Steps 50 -->
|
| 264 |
+
<div class="steps-group">
|
| 265 |
+
<h2 class="title is-3">Flow Matching Steps: 50</h2>
|
| 266 |
+
<div class="experiment-grid" id="smolvla-steps-50"></div>
|
| 267 |
+
</div>
|
| 268 |
|
| 269 |
+
<!-- Steps 100 -->
|
| 270 |
+
<div class="steps-group">
|
| 271 |
+
<h2 class="title is-3">Flow Matching Steps: 100</h2>
|
| 272 |
+
<div class="experiment-grid" id="smolvla-steps-100"></div>
|
| 273 |
+
</div>
|
| 274 |
+
</div>
|
|
|
|
|
|
|
|
|
|
| 275 |
</div>
|
| 276 |
</section>
|
| 277 |
|
| 278 |
+
<!-- Image Modal -->
|
| 279 |
+
<div id="imageModal" class="image-modal" onclick="closeModal()">
|
| 280 |
+
<span class="close-modal">×</span>
|
| 281 |
+
<img id="modalImage" src="">
|
| 282 |
+
</div>
|
| 283 |
|
| 284 |
<footer class="footer">
|
| 285 |
<div class="container">
|
| 286 |
<div class="content has-text-centered">
|
| 287 |
+
<p>
|
| 288 |
+
RTC Experiments for LeRobot - Real-Time Chunking Integration
|
| 289 |
+
</p>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 290 |
</div>
|
| 291 |
</div>
|
| 292 |
</footer>
|
| 293 |
|
| 294 |
+
<script>
|
| 295 |
+
// Experiment data structure
|
| 296 |
+
const experiments = {
|
| 297 |
+
pi05: {
|
| 298 |
+
2: ['0.1', '0.2', '0.5', '0.8', '1.0', '1.2', '1.5'],
|
| 299 |
+
5: ['0.1', '0.2', '0.5', '0.8', '1.0'],
|
| 300 |
+
10: ['0.1', '0.2', '0.5']
|
| 301 |
+
},
|
| 302 |
+
smolvla: {
|
| 303 |
+
2: ['0.1', '0.2', '0.5', '0.8', '1.0', '1.2', '1.5'],
|
| 304 |
+
5: ['0.1', '0.2', '0.5', '0.8', '1.0', '1.2', '1.5'],
|
| 305 |
+
10: ['0.1', '0.2', '0.5', '0.8', '1.0', '1.2', '1.5'],
|
| 306 |
+
20: ['0.1', '0.2', '0.5', '0.8', '1.0', '1.2', '1.5'],
|
| 307 |
+
50: ['0.1', '0.2', '0.5', '0.8', '1.0', '1.2', '1.5'],
|
| 308 |
+
100: ['0.1', '0.2', '0.5', '0.8', '1.0', '1.2', '1.5']
|
| 309 |
+
}
|
| 310 |
+
};
|
| 311 |
+
|
| 312 |
+
const visualizations = {
|
| 313 |
+
'final': 'final_actions_comparison.png',
|
| 314 |
+
'xt': 'denoising_xt_comparison.png',
|
| 315 |
+
'vt': 'denoising_vt_comparison.png',
|
| 316 |
+
'x1t': 'denoising_x1t_comparison.png',
|
| 317 |
+
'correction': 'denoising_correction_comparison.png'
|
| 318 |
+
};
|
| 319 |
+
|
| 320 |
+
const visualizationNames = {
|
| 321 |
+
'final': 'Final Actions',
|
| 322 |
+
'xt': 'X_t',
|
| 323 |
+
'vt': 'V_t (Velocity)',
|
| 324 |
+
'x1t': 'X1_t',
|
| 325 |
+
'correction': 'Correction'
|
| 326 |
+
};
|
| 327 |
+
|
| 328 |
+
// Generate experiment cards
|
| 329 |
+
function generateExperimentCards() {
|
| 330 |
+
for (const model in experiments) {
|
| 331 |
+
for (const steps in experiments[model]) {
|
| 332 |
+
const sigmas = experiments[model][steps];
|
| 333 |
+
const gridId = `${model}-steps-${steps}`;
|
| 334 |
+
const grid = document.getElementById(gridId);
|
| 335 |
+
|
| 336 |
+
if (!grid) continue;
|
| 337 |
+
|
| 338 |
+
sigmas.forEach(sigma => {
|
| 339 |
+
const card = createExperimentCard(model, steps, sigma);
|
| 340 |
+
grid.appendChild(card);
|
| 341 |
+
});
|
| 342 |
+
}
|
| 343 |
+
}
|
| 344 |
+
}
|
| 345 |
+
|
| 346 |
+
function createExperimentCard(model, steps, sigma) {
|
| 347 |
+
const card = document.createElement('div');
|
| 348 |
+
card.className = 'experiment-card';
|
| 349 |
+
|
| 350 |
+
const title = document.createElement('h3');
|
| 351 |
+
title.textContent = `Sigma_d: ${sigma}`;
|
| 352 |
+
card.appendChild(title);
|
| 353 |
+
|
| 354 |
+
// Create visualization tabs
|
| 355 |
+
const tabs = document.createElement('div');
|
| 356 |
+
tabs.className = 'visualization-tabs';
|
| 357 |
+
|
| 358 |
+
let firstTab = true;
|
| 359 |
+
for (const vizKey in visualizations) {
|
| 360 |
+
const tab = document.createElement('div');
|
| 361 |
+
tab.className = 'visualization-tab' + (firstTab ? ' active' : '');
|
| 362 |
+
tab.textContent = visualizationNames[vizKey];
|
| 363 |
+
tab.onclick = () => switchVisualization(card, model, steps, sigma, vizKey);
|
| 364 |
+
tabs.appendChild(tab);
|
| 365 |
+
firstTab = false;
|
| 366 |
+
}
|
| 367 |
+
card.appendChild(tabs);
|
| 368 |
+
|
| 369 |
+
// Create image element
|
| 370 |
+
const img = document.createElement('img');
|
| 371 |
+
const expPath = `./static/experiments/${model}_steps_${steps}_sigma_${sigma}`;
|
| 372 |
+
img.src = `${expPath}/${visualizations['final']}`;
|
| 373 |
+
img.alt = `${model} - Steps: ${steps}, Sigma: ${sigma}`;
|
| 374 |
+
img.onclick = () => openModal(img.src);
|
| 375 |
+
card.appendChild(img);
|
| 376 |
+
|
| 377 |
+
return card;
|
| 378 |
+
}
|
| 379 |
+
|
| 380 |
+
function switchVisualization(card, model, steps, sigma, vizKey) {
|
| 381 |
+
// Update active tab
|
| 382 |
+
const tabs = card.querySelectorAll('.visualization-tab');
|
| 383 |
+
tabs.forEach(tab => {
|
| 384 |
+
tab.classList.remove('active');
|
| 385 |
+
if (tab.textContent === visualizationNames[vizKey]) {
|
| 386 |
+
tab.classList.add('active');
|
| 387 |
+
}
|
| 388 |
+
});
|
| 389 |
+
|
| 390 |
+
// Update image
|
| 391 |
+
const img = card.querySelector('img');
|
| 392 |
+
const expPath = `./static/experiments/${model}_steps_${steps}_sigma_${sigma}`;
|
| 393 |
+
img.src = `${expPath}/${visualizations[vizKey]}`;
|
| 394 |
+
}
|
| 395 |
+
|
| 396 |
+
function showModel(model) {
|
| 397 |
+
// Update nav tabs
|
| 398 |
+
document.querySelectorAll('.nav-tab').forEach(tab => {
|
| 399 |
+
tab.classList.remove('active');
|
| 400 |
+
if (tab.textContent.toLowerCase().includes(model.replace('smolvla', 'smol'))) {
|
| 401 |
+
tab.classList.add('active');
|
| 402 |
+
}
|
| 403 |
+
});
|
| 404 |
+
|
| 405 |
+
// Show/hide sections
|
| 406 |
+
document.getElementById('pi05-section').style.display = model === 'pi05' ? 'block' : 'none';
|
| 407 |
+
document.getElementById('smolvla-section').style.display = model === 'smolvla' ? 'block' : 'none';
|
| 408 |
+
}
|
| 409 |
+
|
| 410 |
+
function openModal(src) {
|
| 411 |
+
const modal = document.getElementById('imageModal');
|
| 412 |
+
const modalImg = document.getElementById('modalImage');
|
| 413 |
+
modal.style.display = 'block';
|
| 414 |
+
modalImg.src = src;
|
| 415 |
+
}
|
| 416 |
+
|
| 417 |
+
function closeModal() {
|
| 418 |
+
document.getElementById('imageModal').style.display = 'none';
|
| 419 |
+
}
|
| 420 |
+
|
| 421 |
+
// Close modal with Escape key
|
| 422 |
+
document.addEventListener('keydown', function(event) {
|
| 423 |
+
if (event.key === 'Escape') {
|
| 424 |
+
closeModal();
|
| 425 |
+
}
|
| 426 |
+
});
|
| 427 |
+
|
| 428 |
+
// Initialize
|
| 429 |
+
generateExperimentCards();
|
| 430 |
+
</script>
|
| 431 |
+
|
| 432 |
</body>
|
| 433 |
</html>
|
static/experiments/experiment_log.txt
ADDED
|
@@ -0,0 +1,378 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
RTC Evaluation Experiments - Sat Nov 22 23:39:40 +07 2025
|
| 2 |
+
======================================
|
| 3 |
+
Sat Nov 22 23:39:40 +07 2025: Starting pi05 steps=10 sigma_d=0.1
|
| 4 |
+
WARNING:lerobot.configs.policies:Device 'cuda' is not available. Switching to 'mps'.
|
| 5 |
+
WARNING:lerobot.configs.policies:Device 'cuda' is not available. Switching to 'mps'.
|
| 6 |
+
INFO 2025-11-22 23:39:46 _dataset.py:959 ================================================================================
|
| 7 |
+
INFO 2025-11-22 23:39:46 _dataset.py:960 RTC Dataset Evaluation
|
| 8 |
+
INFO 2025-11-22 23:39:46 _dataset.py:961 Config: RTCEvalConfig(policy=PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path='lerobot/pi05_libero_finetuned', paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=None, image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=True, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06), dataset=DatasetConfig(repo_id='HuggingFaceVLA/libero', root=None, episodes=None, image_transforms=ImageTransformsConfig(enable=False, max_num_transforms=3, random_order=False, tfs={'brightness': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'brightness': (0.8, 1.2)}), 'contrast': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'contrast': (0.8, 1.2)}), 'saturation': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'saturation': (0.5, 1.5)}), 'hue': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'hue': (-0.05, 0.05)}), 'sharpness': ImageTransformConfig(weight=1.0, type='SharpnessJitter', kwargs={'sharpness': (0.5, 1.5)}), 'affine': ImageTransformConfig(weight=1.0, type='RandomAffine', kwargs={'degrees': (-5.0, 5.0), 'translate': (0.05, 0.05)})}), revision=None, use_imagenet_stats=True, video_backend='torchcodec', streaming=False), rtc=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=None, execution_horizon=8, sigma_d=0.1, debug=False, debug_maxlen=100), device='mps', output_dir='rtc_experiments_results/pi05_steps_10_sigma_0.1', seed=10, inference_delay=4, num_inference_steps=10, use_torch_compile=False, torch_compile_backend='inductor', torch_compile_mode='default', torch_compile_disable_cudagraphs=True)
|
| 9 |
+
INFO 2025-11-22 23:39:46 _dataset.py:962 ================================================================================
|
| 10 |
+
INFO 2025-11-22 23:39:46 _dataset.py:256 Loading dataset: HuggingFaceVLA/libero
|
| 11 |
+
INFO 2025-11-22 23:39:46 _dataset.py:269 Dataset loaded: 273465 samples, 1693 episodes
|
| 12 |
+
INFO 2025-11-22 23:39:49 _dataset.py:280 ================================================================================
|
| 13 |
+
INFO 2025-11-22 23:39:49 _dataset.py:281 Ready to run evaluation with sequential policy loading:
|
| 14 |
+
INFO 2025-11-22 23:39:49 _dataset.py:282 1. policy_prev_chunk - Generate reference chunk, then destroy
|
| 15 |
+
INFO 2025-11-22 23:39:49 _dataset.py:283 2. policy_no_rtc - Generate without RTC, then destroy
|
| 16 |
+
INFO 2025-11-22 23:39:49 _dataset.py:284 3. policy_rtc - Generate with RTC, then destroy
|
| 17 |
+
INFO 2025-11-22 23:39:49 _dataset.py:285 Note: Only one policy in memory at a time for efficient memory usage
|
| 18 |
+
INFO 2025-11-22 23:39:49 _dataset.py:286 ================================================================================
|
| 19 |
+
INFO 2025-11-22 23:39:49 _dataset.py:445 Output directory: rtc_experiments_results/pi05_steps_10_sigma_0.1
|
| 20 |
+
INFO 2025-11-22 23:39:49 _dataset.py:447 ================================================================================
|
| 21 |
+
INFO 2025-11-22 23:39:49 _dataset.py:448 Starting RTC evaluation
|
| 22 |
+
INFO 2025-11-22 23:39:49 _dataset.py:449 Inference delay: 4
|
| 23 |
+
INFO 2025-11-22 23:39:49 _dataset.py:451 Number of flow matching steps: 10
|
| 24 |
+
INFO 2025-11-22 23:39:49 _dataset.py:454 ================================================================================
|
| 25 |
+
INFO 2025-11-22 23:39:49 _dataset.py:469 ================================================================================
|
| 26 |
+
INFO 2025-11-22 23:39:49 _dataset.py:470 Step 1: Generating previous chunk with policy_prev_chunk
|
| 27 |
+
INFO 2025-11-22 23:39:49 _dataset.py:471 ================================================================================
|
| 28 |
+
INFO 2025-11-22 23:39:49 _dataset.py:299 Initializing policy_prev_chunk...
|
| 29 |
+
INFO 2025-11-22 23:39:49 ils/utils.py:46 Metal backend detected, using mps.
|
| 30 |
+
WARNING 2025-11-22 23:39:49 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 31 |
+
INFO 2025-11-22 23:39:49 ils/utils.py:46 Metal backend detected, using mps.
|
| 32 |
+
WARNING 2025-11-22 23:39:49 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 33 |
+
INFO 2025-11-22 23:39:49 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 34 |
+
INFO 2025-11-22 23:40:58 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 35 |
+
WARNING 2025-11-22 23:41:01 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 36 |
+
WARNING 2025-11-22 23:41:01 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 37 |
+
INFO 2025-11-22 23:41:02 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 38 |
+
INFO 2025-11-22 23:41:02 _dataset.py:336 RTC enabled: False
|
| 39 |
+
INFO 2025-11-22 23:41:02 _dataset.py:337 RTC debug: False
|
| 40 |
+
INFO 2025-11-22 23:41:02 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=False, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 41 |
+
INFO 2025-11-22 23:41:02 _dataset.py:344 ✓ policy_prev_chunk initialized successfully
|
| 42 |
+
INFO 2025-11-22 23:41:04 _dataset.py:483 Generated prev_chunk shape: torch.Size([25, 7])
|
| 43 |
+
INFO 2025-11-22 23:41:04 _dataset.py:410 Destroying policy_prev_chunk and freeing memory...
|
| 44 |
+
INFO 2025-11-22 23:41:05 _dataset.py:430 ✓ policy_prev_chunk destroyed and memory freed
|
| 45 |
+
INFO 2025-11-22 23:41:05 _dataset.py:491 ================================================================================
|
| 46 |
+
INFO 2025-11-22 23:41:05 _dataset.py:492 Step 2: Generating actions WITHOUT RTC with policy_no_rtc
|
| 47 |
+
INFO 2025-11-22 23:41:05 _dataset.py:493 ================================================================================
|
| 48 |
+
INFO 2025-11-22 23:41:05 _dataset.py:299 Initializing policy_no_rtc...
|
| 49 |
+
INFO 2025-11-22 23:41:07 ils/utils.py:46 Metal backend detected, using mps.
|
| 50 |
+
WARNING 2025-11-22 23:41:07 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 51 |
+
INFO 2025-11-22 23:41:07 ils/utils.py:46 Metal backend detected, using mps.
|
| 52 |
+
WARNING 2025-11-22 23:41:07 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 53 |
+
INFO 2025-11-22 23:41:07 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 54 |
+
INFO 2025-11-22 23:42:17 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 55 |
+
WARNING 2025-11-22 23:42:19 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 56 |
+
WARNING 2025-11-22 23:42:19 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 57 |
+
INFO 2025-11-22 23:42:25 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 58 |
+
INFO 2025-11-22 23:42:25 _dataset.py:336 RTC enabled: False
|
| 59 |
+
INFO 2025-11-22 23:42:25 _dataset.py:337 RTC debug: True
|
| 60 |
+
INFO 2025-11-22 23:42:25 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=True, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 61 |
+
INFO 2025-11-22 23:42:25 _dataset.py:344 ✓ policy_no_rtc initialized successfully
|
| 62 |
+
INFO 2025-11-22 23:42:28 _dataset.py:515 Tracked 10 steps without RTC
|
| 63 |
+
INFO 2025-11-22 23:42:28 _dataset.py:516 Generated no_rtc_actions shape: torch.Size([1, 50, 7])
|
| 64 |
+
INFO 2025-11-22 23:42:28 _dataset.py:410 Destroying policy_no_rtc and freeing memory...
|
| 65 |
+
INFO 2025-11-22 23:42:30 _dataset.py:430 ✓ policy_no_rtc destroyed and memory freed
|
| 66 |
+
INFO 2025-11-22 23:42:30 _dataset.py:524 ================================================================================
|
| 67 |
+
INFO 2025-11-22 23:42:30 _dataset.py:525 Step 3: Generating actions WITH RTC with policy_rtc
|
| 68 |
+
INFO 2025-11-22 23:42:30 _dataset.py:526 ================================================================================
|
| 69 |
+
INFO 2025-11-22 23:42:30 _dataset.py:299 Initializing policy_rtc...
|
| 70 |
+
INFO 2025-11-22 23:42:31 ils/utils.py:46 Metal backend detected, using mps.
|
| 71 |
+
WARNING 2025-11-22 23:42:31 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 72 |
+
INFO 2025-11-22 23:42:31 ils/utils.py:46 Metal backend detected, using mps.
|
| 73 |
+
WARNING 2025-11-22 23:42:31 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 74 |
+
INFO 2025-11-22 23:42:31 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 75 |
+
INFO 2025-11-22 23:43:39 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 76 |
+
WARNING 2025-11-22 23:43:41 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 77 |
+
WARNING 2025-11-22 23:43:41 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 78 |
+
INFO 2025-11-22 23:43:56 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 79 |
+
INFO 2025-11-22 23:43:56 _dataset.py:336 RTC enabled: True
|
| 80 |
+
INFO 2025-11-22 23:43:56 _dataset.py:337 RTC debug: True
|
| 81 |
+
INFO 2025-11-22 23:43:56 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=True, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=True, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 82 |
+
INFO 2025-11-22 23:43:56 _dataset.py:344 ✓ policy_rtc initialized successfully
|
| 83 |
+
INFO 2025-11-22 23:43:59 _dataset.py:546 Tracked 10 steps with RTC
|
| 84 |
+
INFO 2025-11-22 23:43:59 _dataset.py:547 Generated rtc_actions shape: torch.Size([1, 50, 7])
|
| 85 |
+
ERROR 2025-11-22 23:43:59 _dataset.py:553 Error getting num_steps: 'PI05Config' object has no attribute 'num_steps'
|
| 86 |
+
WARNING 2025-11-22 23:43:59 _dataset.py:555 Using num_inference_steps: 10 instead of num_steps
|
| 87 |
+
INFO 2025-11-22 23:43:59 _dataset.py:410 Destroying policy_rtc and freeing memory...
|
| 88 |
+
INFO 2025-11-22 23:44:01 _dataset.py:430 ✓ policy_rtc destroyed and memory freed
|
| 89 |
+
INFO 2025-11-22 23:44:01 _dataset.py:561 ================================================================================
|
| 90 |
+
INFO 2025-11-22 23:44:01 _dataset.py:562 Plotting results...
|
| 91 |
+
INFO 2025-11-22 23:44:01 _dataset.py:818 ================================================================================
|
| 92 |
+
INFO 2025-11-22 23:44:01 _dataset.py:819 Plotting 10 steps
|
| 93 |
+
INFO 2025-11-22 23:44:02 _dataset.py:818 ================================================================================
|
| 94 |
+
INFO 2025-11-22 23:44:02 _dataset.py:819 Plotting 10 steps
|
| 95 |
+
INFO 2025-11-22 23:44:02 _dataset.py:905 Plotting final no-RTC x_t step as orange dashed reference
|
| 96 |
+
INFO 2025-11-22 23:44:03 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.1/denoising_xt_comparison.png
|
| 97 |
+
INFO 2025-11-22 23:44:03 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.1/denoising_vt_comparison.png
|
| 98 |
+
INFO 2025-11-22 23:44:04 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.1/denoising_correction_comparison.png
|
| 99 |
+
INFO 2025-11-22 23:44:05 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.1/denoising_x1t_comparison.png
|
| 100 |
+
INFO 2025-11-22 23:44:05 _dataset.py:566 ================================================================================
|
| 101 |
+
INFO 2025-11-22 23:44:05 _dataset.py:567 Plotting final actions comparison...
|
| 102 |
+
INFO 2025-11-22 23:44:05 _dataset.py:687 Saved final actions comparison to rtc_experiments_results/pi05_steps_10_sigma_0.1/final_actions_comparison.png
|
| 103 |
+
INFO 2025-11-22 23:44:05 _dataset.py:570 ================================================================================
|
| 104 |
+
INFO 2025-11-22 23:44:05 _dataset.py:571 Evaluation completed successfully
|
| 105 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 106 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 107 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 108 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 109 |
+
✓ Loaded state dict from model.safetensors
|
| 110 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 111 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 112 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 113 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 114 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 115 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 116 |
+
✓ Loaded state dict from model.safetensors
|
| 117 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 118 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 119 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 120 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 121 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 122 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 123 |
+
✓ Loaded state dict from model.safetensors
|
| 124 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 125 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 126 |
+
Sat Nov 22 23:44:08 +07 2025: SUCCESS pi05 steps=10 sigma_d=0.1
|
| 127 |
+
|
| 128 |
+
Sat Nov 22 23:44:08 +07 2025: Starting pi05 steps=10 sigma_d=0.2
|
| 129 |
+
WARNING:lerobot.configs.policies:Device 'cuda' is not available. Switching to 'mps'.
|
| 130 |
+
WARNING:lerobot.configs.policies:Device 'cuda' is not available. Switching to 'mps'.
|
| 131 |
+
INFO 2025-11-22 23:44:14 _dataset.py:959 ================================================================================
|
| 132 |
+
INFO 2025-11-22 23:44:14 _dataset.py:960 RTC Dataset Evaluation
|
| 133 |
+
INFO 2025-11-22 23:44:14 _dataset.py:961 Config: RTCEvalConfig(policy=PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path='lerobot/pi05_libero_finetuned', paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=None, image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=True, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06), dataset=DatasetConfig(repo_id='HuggingFaceVLA/libero', root=None, episodes=None, image_transforms=ImageTransformsConfig(enable=False, max_num_transforms=3, random_order=False, tfs={'brightness': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'brightness': (0.8, 1.2)}), 'contrast': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'contrast': (0.8, 1.2)}), 'saturation': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'saturation': (0.5, 1.5)}), 'hue': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'hue': (-0.05, 0.05)}), 'sharpness': ImageTransformConfig(weight=1.0, type='SharpnessJitter', kwargs={'sharpness': (0.5, 1.5)}), 'affine': ImageTransformConfig(weight=1.0, type='RandomAffine', kwargs={'degrees': (-5.0, 5.0), 'translate': (0.05, 0.05)})}), revision=None, use_imagenet_stats=True, video_backend='torchcodec', streaming=False), rtc=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=None, execution_horizon=8, sigma_d=0.2, debug=False, debug_maxlen=100), device='mps', output_dir='rtc_experiments_results/pi05_steps_10_sigma_0.2', seed=10, inference_delay=4, num_inference_steps=10, use_torch_compile=False, torch_compile_backend='inductor', torch_compile_mode='default', torch_compile_disable_cudagraphs=True)
|
| 134 |
+
INFO 2025-11-22 23:44:14 _dataset.py:962 ================================================================================
|
| 135 |
+
INFO 2025-11-22 23:44:14 _dataset.py:256 Loading dataset: HuggingFaceVLA/libero
|
| 136 |
+
INFO 2025-11-22 23:44:15 _dataset.py:269 Dataset loaded: 273465 samples, 1693 episodes
|
| 137 |
+
INFO 2025-11-22 23:44:17 _dataset.py:280 ================================================================================
|
| 138 |
+
INFO 2025-11-22 23:44:17 _dataset.py:281 Ready to run evaluation with sequential policy loading:
|
| 139 |
+
INFO 2025-11-22 23:44:17 _dataset.py:282 1. policy_prev_chunk - Generate reference chunk, then destroy
|
| 140 |
+
INFO 2025-11-22 23:44:17 _dataset.py:283 2. policy_no_rtc - Generate without RTC, then destroy
|
| 141 |
+
INFO 2025-11-22 23:44:17 _dataset.py:284 3. policy_rtc - Generate with RTC, then destroy
|
| 142 |
+
INFO 2025-11-22 23:44:17 _dataset.py:285 Note: Only one policy in memory at a time for efficient memory usage
|
| 143 |
+
INFO 2025-11-22 23:44:17 _dataset.py:286 ================================================================================
|
| 144 |
+
INFO 2025-11-22 23:44:17 _dataset.py:445 Output directory: rtc_experiments_results/pi05_steps_10_sigma_0.2
|
| 145 |
+
INFO 2025-11-22 23:44:17 _dataset.py:447 ================================================================================
|
| 146 |
+
INFO 2025-11-22 23:44:17 _dataset.py:448 Starting RTC evaluation
|
| 147 |
+
INFO 2025-11-22 23:44:17 _dataset.py:449 Inference delay: 4
|
| 148 |
+
INFO 2025-11-22 23:44:17 _dataset.py:451 Number of flow matching steps: 10
|
| 149 |
+
INFO 2025-11-22 23:44:17 _dataset.py:454 ================================================================================
|
| 150 |
+
INFO 2025-11-22 23:44:17 _dataset.py:469 ================================================================================
|
| 151 |
+
INFO 2025-11-22 23:44:17 _dataset.py:470 Step 1: Generating previous chunk with policy_prev_chunk
|
| 152 |
+
INFO 2025-11-22 23:44:17 _dataset.py:471 ================================================================================
|
| 153 |
+
INFO 2025-11-22 23:44:17 _dataset.py:299 Initializing policy_prev_chunk...
|
| 154 |
+
INFO 2025-11-22 23:44:17 ils/utils.py:46 Metal backend detected, using mps.
|
| 155 |
+
WARNING 2025-11-22 23:44:17 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 156 |
+
INFO 2025-11-22 23:44:17 ils/utils.py:46 Metal backend detected, using mps.
|
| 157 |
+
WARNING 2025-11-22 23:44:17 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 158 |
+
INFO 2025-11-22 23:44:17 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 159 |
+
INFO 2025-11-22 23:45:26 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 160 |
+
WARNING 2025-11-22 23:45:28 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 161 |
+
WARNING 2025-11-22 23:45:28 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 162 |
+
INFO 2025-11-22 23:45:31 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 163 |
+
INFO 2025-11-22 23:45:31 _dataset.py:336 RTC enabled: False
|
| 164 |
+
INFO 2025-11-22 23:45:31 _dataset.py:337 RTC debug: False
|
| 165 |
+
INFO 2025-11-22 23:45:31 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=False, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 166 |
+
INFO 2025-11-22 23:45:31 _dataset.py:344 ✓ policy_prev_chunk initialized successfully
|
| 167 |
+
INFO 2025-11-22 23:45:33 _dataset.py:483 Generated prev_chunk shape: torch.Size([25, 7])
|
| 168 |
+
INFO 2025-11-22 23:45:33 _dataset.py:410 Destroying policy_prev_chunk and freeing memory...
|
| 169 |
+
INFO 2025-11-22 23:45:34 _dataset.py:430 ✓ policy_prev_chunk destroyed and memory freed
|
| 170 |
+
INFO 2025-11-22 23:45:34 _dataset.py:491 ================================================================================
|
| 171 |
+
INFO 2025-11-22 23:45:34 _dataset.py:492 Step 2: Generating actions WITHOUT RTC with policy_no_rtc
|
| 172 |
+
INFO 2025-11-22 23:45:34 _dataset.py:493 ================================================================================
|
| 173 |
+
INFO 2025-11-22 23:45:34 _dataset.py:299 Initializing policy_no_rtc...
|
| 174 |
+
INFO 2025-11-22 23:45:34 ils/utils.py:46 Metal backend detected, using mps.
|
| 175 |
+
WARNING 2025-11-22 23:45:34 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 176 |
+
INFO 2025-11-22 23:45:35 ils/utils.py:46 Metal backend detected, using mps.
|
| 177 |
+
WARNING 2025-11-22 23:45:35 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 178 |
+
INFO 2025-11-22 23:45:35 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 179 |
+
INFO 2025-11-22 23:46:43 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 180 |
+
WARNING 2025-11-22 23:46:44 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 181 |
+
WARNING 2025-11-22 23:46:44 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 182 |
+
INFO 2025-11-22 23:46:46 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 183 |
+
INFO 2025-11-22 23:46:46 _dataset.py:336 RTC enabled: False
|
| 184 |
+
INFO 2025-11-22 23:46:46 _dataset.py:337 RTC debug: True
|
| 185 |
+
INFO 2025-11-22 23:46:46 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=True, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 186 |
+
INFO 2025-11-22 23:46:46 _dataset.py:344 ✓ policy_no_rtc initialized successfully
|
| 187 |
+
INFO 2025-11-22 23:46:48 _dataset.py:515 Tracked 10 steps without RTC
|
| 188 |
+
INFO 2025-11-22 23:46:48 _dataset.py:516 Generated no_rtc_actions shape: torch.Size([1, 50, 7])
|
| 189 |
+
INFO 2025-11-22 23:46:48 _dataset.py:410 Destroying policy_no_rtc and freeing memory...
|
| 190 |
+
INFO 2025-11-22 23:46:49 _dataset.py:430 ✓ policy_no_rtc destroyed and memory freed
|
| 191 |
+
INFO 2025-11-22 23:46:49 _dataset.py:524 ================================================================================
|
| 192 |
+
INFO 2025-11-22 23:46:49 _dataset.py:525 Step 3: Generating actions WITH RTC with policy_rtc
|
| 193 |
+
INFO 2025-11-22 23:46:49 _dataset.py:526 ================================================================================
|
| 194 |
+
INFO 2025-11-22 23:46:49 _dataset.py:299 Initializing policy_rtc...
|
| 195 |
+
INFO 2025-11-22 23:46:49 ils/utils.py:46 Metal backend detected, using mps.
|
| 196 |
+
WARNING 2025-11-22 23:46:49 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 197 |
+
INFO 2025-11-22 23:46:49 ils/utils.py:46 Metal backend detected, using mps.
|
| 198 |
+
WARNING 2025-11-22 23:46:49 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 199 |
+
INFO 2025-11-22 23:46:49 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 200 |
+
INFO 2025-11-22 23:47:56 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 201 |
+
WARNING 2025-11-22 23:47:58 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 202 |
+
WARNING 2025-11-22 23:47:58 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 203 |
+
INFO 2025-11-22 23:48:07 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 204 |
+
INFO 2025-11-22 23:48:07 _dataset.py:336 RTC enabled: True
|
| 205 |
+
INFO 2025-11-22 23:48:07 _dataset.py:337 RTC debug: True
|
| 206 |
+
INFO 2025-11-22 23:48:07 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=True, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=True, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 207 |
+
INFO 2025-11-22 23:48:07 _dataset.py:344 ✓ policy_rtc initialized successfully
|
| 208 |
+
INFO 2025-11-22 23:48:10 _dataset.py:546 Tracked 10 steps with RTC
|
| 209 |
+
INFO 2025-11-22 23:48:10 _dataset.py:547 Generated rtc_actions shape: torch.Size([1, 50, 7])
|
| 210 |
+
ERROR 2025-11-22 23:48:10 _dataset.py:553 Error getting num_steps: 'PI05Config' object has no attribute 'num_steps'
|
| 211 |
+
WARNING 2025-11-22 23:48:10 _dataset.py:555 Using num_inference_steps: 10 instead of num_steps
|
| 212 |
+
INFO 2025-11-22 23:48:10 _dataset.py:410 Destroying policy_rtc and freeing memory...
|
| 213 |
+
INFO 2025-11-22 23:48:12 _dataset.py:430 ✓ policy_rtc destroyed and memory freed
|
| 214 |
+
INFO 2025-11-22 23:48:12 _dataset.py:561 ================================================================================
|
| 215 |
+
INFO 2025-11-22 23:48:12 _dataset.py:562 Plotting results...
|
| 216 |
+
INFO 2025-11-22 23:48:12 _dataset.py:818 ================================================================================
|
| 217 |
+
INFO 2025-11-22 23:48:12 _dataset.py:819 Plotting 10 steps
|
| 218 |
+
INFO 2025-11-22 23:48:13 _dataset.py:818 ================================================================================
|
| 219 |
+
INFO 2025-11-22 23:48:13 _dataset.py:819 Plotting 10 steps
|
| 220 |
+
INFO 2025-11-22 23:48:13 _dataset.py:905 Plotting final no-RTC x_t step as orange dashed reference
|
| 221 |
+
INFO 2025-11-22 23:48:14 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.2/denoising_xt_comparison.png
|
| 222 |
+
INFO 2025-11-22 23:48:15 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.2/denoising_vt_comparison.png
|
| 223 |
+
INFO 2025-11-22 23:48:15 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.2/denoising_correction_comparison.png
|
| 224 |
+
INFO 2025-11-22 23:48:16 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.2/denoising_x1t_comparison.png
|
| 225 |
+
INFO 2025-11-22 23:48:16 _dataset.py:566 ================================================================================
|
| 226 |
+
INFO 2025-11-22 23:48:16 _dataset.py:567 Plotting final actions comparison...
|
| 227 |
+
INFO 2025-11-22 23:48:17 _dataset.py:687 Saved final actions comparison to rtc_experiments_results/pi05_steps_10_sigma_0.2/final_actions_comparison.png
|
| 228 |
+
INFO 2025-11-22 23:48:17 _dataset.py:570 ================================================================================
|
| 229 |
+
INFO 2025-11-22 23:48:17 _dataset.py:571 Evaluation completed successfully
|
| 230 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 231 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 232 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 233 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 234 |
+
✓ Loaded state dict from model.safetensors
|
| 235 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 236 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 237 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 238 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 239 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 240 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 241 |
+
✓ Loaded state dict from model.safetensors
|
| 242 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 243 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 244 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 245 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 246 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 247 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 248 |
+
✓ Loaded state dict from model.safetensors
|
| 249 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 250 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 251 |
+
Sat Nov 22 23:48:19 +07 2025: SUCCESS pi05 steps=10 sigma_d=0.2
|
| 252 |
+
|
| 253 |
+
Sat Nov 22 23:48:19 +07 2025: Starting pi05 steps=10 sigma_d=0.5
|
| 254 |
+
WARNING:lerobot.configs.policies:Device 'cuda' is not available. Switching to 'mps'.
|
| 255 |
+
WARNING:lerobot.configs.policies:Device 'cuda' is not available. Switching to 'mps'.
|
| 256 |
+
INFO 2025-11-22 23:48:25 _dataset.py:959 ================================================================================
|
| 257 |
+
INFO 2025-11-22 23:48:25 _dataset.py:960 RTC Dataset Evaluation
|
| 258 |
+
INFO 2025-11-22 23:48:25 _dataset.py:961 Config: RTCEvalConfig(policy=PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path='lerobot/pi05_libero_finetuned', paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=None, image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=True, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06), dataset=DatasetConfig(repo_id='HuggingFaceVLA/libero', root=None, episodes=None, image_transforms=ImageTransformsConfig(enable=False, max_num_transforms=3, random_order=False, tfs={'brightness': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'brightness': (0.8, 1.2)}), 'contrast': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'contrast': (0.8, 1.2)}), 'saturation': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'saturation': (0.5, 1.5)}), 'hue': ImageTransformConfig(weight=1.0, type='ColorJitter', kwargs={'hue': (-0.05, 0.05)}), 'sharpness': ImageTransformConfig(weight=1.0, type='SharpnessJitter', kwargs={'sharpness': (0.5, 1.5)}), 'affine': ImageTransformConfig(weight=1.0, type='RandomAffine', kwargs={'degrees': (-5.0, 5.0), 'translate': (0.05, 0.05)})}), revision=None, use_imagenet_stats=True, video_backend='torchcodec', streaming=False), rtc=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=None, execution_horizon=8, sigma_d=0.5, debug=False, debug_maxlen=100), device='mps', output_dir='rtc_experiments_results/pi05_steps_10_sigma_0.5', seed=10, inference_delay=4, num_inference_steps=10, use_torch_compile=False, torch_compile_backend='inductor', torch_compile_mode='default', torch_compile_disable_cudagraphs=True)
|
| 259 |
+
INFO 2025-11-22 23:48:25 _dataset.py:962 ================================================================================
|
| 260 |
+
INFO 2025-11-22 23:48:25 _dataset.py:256 Loading dataset: HuggingFaceVLA/libero
|
| 261 |
+
INFO 2025-11-22 23:48:26 _dataset.py:269 Dataset loaded: 273465 samples, 1693 episodes
|
| 262 |
+
INFO 2025-11-22 23:48:28 _dataset.py:280 ================================================================================
|
| 263 |
+
INFO 2025-11-22 23:48:28 _dataset.py:281 Ready to run evaluation with sequential policy loading:
|
| 264 |
+
INFO 2025-11-22 23:48:28 _dataset.py:282 1. policy_prev_chunk - Generate reference chunk, then destroy
|
| 265 |
+
INFO 2025-11-22 23:48:28 _dataset.py:283 2. policy_no_rtc - Generate without RTC, then destroy
|
| 266 |
+
INFO 2025-11-22 23:48:28 _dataset.py:284 3. policy_rtc - Generate with RTC, then destroy
|
| 267 |
+
INFO 2025-11-22 23:48:28 _dataset.py:285 Note: Only one policy in memory at a time for efficient memory usage
|
| 268 |
+
INFO 2025-11-22 23:48:28 _dataset.py:286 ================================================================================
|
| 269 |
+
INFO 2025-11-22 23:48:28 _dataset.py:445 Output directory: rtc_experiments_results/pi05_steps_10_sigma_0.5
|
| 270 |
+
INFO 2025-11-22 23:48:28 _dataset.py:447 ================================================================================
|
| 271 |
+
INFO 2025-11-22 23:48:28 _dataset.py:448 Starting RTC evaluation
|
| 272 |
+
INFO 2025-11-22 23:48:28 _dataset.py:449 Inference delay: 4
|
| 273 |
+
INFO 2025-11-22 23:48:28 _dataset.py:451 Number of flow matching steps: 10
|
| 274 |
+
INFO 2025-11-22 23:48:28 _dataset.py:454 ================================================================================
|
| 275 |
+
INFO 2025-11-22 23:48:28 _dataset.py:469 ================================================================================
|
| 276 |
+
INFO 2025-11-22 23:48:28 _dataset.py:470 Step 1: Generating previous chunk with policy_prev_chunk
|
| 277 |
+
INFO 2025-11-22 23:48:28 _dataset.py:471 ================================================================================
|
| 278 |
+
INFO 2025-11-22 23:48:28 _dataset.py:299 Initializing policy_prev_chunk...
|
| 279 |
+
INFO 2025-11-22 23:48:28 ils/utils.py:46 Metal backend detected, using mps.
|
| 280 |
+
WARNING 2025-11-22 23:48:28 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 281 |
+
INFO 2025-11-22 23:48:28 ils/utils.py:46 Metal backend detected, using mps.
|
| 282 |
+
WARNING 2025-11-22 23:48:28 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 283 |
+
INFO 2025-11-22 23:48:28 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 284 |
+
INFO 2025-11-22 23:49:37 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 285 |
+
WARNING 2025-11-22 23:49:39 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 286 |
+
WARNING 2025-11-22 23:49:39 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 287 |
+
INFO 2025-11-22 23:49:42 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 288 |
+
INFO 2025-11-22 23:49:42 _dataset.py:336 RTC enabled: False
|
| 289 |
+
INFO 2025-11-22 23:49:42 _dataset.py:337 RTC debug: False
|
| 290 |
+
INFO 2025-11-22 23:49:42 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=False, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 291 |
+
INFO 2025-11-22 23:49:42 _dataset.py:344 ✓ policy_prev_chunk initialized successfully
|
| 292 |
+
INFO 2025-11-22 23:49:44 _dataset.py:483 Generated prev_chunk shape: torch.Size([25, 7])
|
| 293 |
+
INFO 2025-11-22 23:49:44 _dataset.py:410 Destroying policy_prev_chunk and freeing memory...
|
| 294 |
+
INFO 2025-11-22 23:49:45 _dataset.py:430 ✓ policy_prev_chunk destroyed and memory freed
|
| 295 |
+
INFO 2025-11-22 23:49:45 _dataset.py:491 ================================================================================
|
| 296 |
+
INFO 2025-11-22 23:49:45 _dataset.py:492 Step 2: Generating actions WITHOUT RTC with policy_no_rtc
|
| 297 |
+
INFO 2025-11-22 23:49:45 _dataset.py:493 ================================================================================
|
| 298 |
+
INFO 2025-11-22 23:49:45 _dataset.py:299 Initializing policy_no_rtc...
|
| 299 |
+
INFO 2025-11-22 23:49:46 ils/utils.py:46 Metal backend detected, using mps.
|
| 300 |
+
WARNING 2025-11-22 23:49:46 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 301 |
+
INFO 2025-11-22 23:49:46 ils/utils.py:46 Metal backend detected, using mps.
|
| 302 |
+
WARNING 2025-11-22 23:49:46 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 303 |
+
INFO 2025-11-22 23:49:46 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 304 |
+
INFO 2025-11-22 23:50:55 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 305 |
+
WARNING 2025-11-22 23:50:57 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 306 |
+
WARNING 2025-11-22 23:50:57 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 307 |
+
INFO 2025-11-22 23:50:58 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 308 |
+
INFO 2025-11-22 23:50:58 _dataset.py:336 RTC enabled: False
|
| 309 |
+
INFO 2025-11-22 23:50:58 _dataset.py:337 RTC debug: True
|
| 310 |
+
INFO 2025-11-22 23:50:58 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=False, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=True, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 311 |
+
INFO 2025-11-22 23:50:58 _dataset.py:344 ✓ policy_no_rtc initialized successfully
|
| 312 |
+
INFO 2025-11-22 23:51:00 _dataset.py:515 Tracked 10 steps without RTC
|
| 313 |
+
INFO 2025-11-22 23:51:00 _dataset.py:516 Generated no_rtc_actions shape: torch.Size([1, 50, 7])
|
| 314 |
+
INFO 2025-11-22 23:51:00 _dataset.py:410 Destroying policy_no_rtc and freeing memory...
|
| 315 |
+
INFO 2025-11-22 23:51:01 _dataset.py:430 ✓ policy_no_rtc destroyed and memory freed
|
| 316 |
+
INFO 2025-11-22 23:51:01 _dataset.py:524 ================================================================================
|
| 317 |
+
INFO 2025-11-22 23:51:01 _dataset.py:525 Step 3: Generating actions WITH RTC with policy_rtc
|
| 318 |
+
INFO 2025-11-22 23:51:01 _dataset.py:526 ================================================================================
|
| 319 |
+
INFO 2025-11-22 23:51:01 _dataset.py:299 Initializing policy_rtc...
|
| 320 |
+
INFO 2025-11-22 23:51:01 ils/utils.py:46 Metal backend detected, using mps.
|
| 321 |
+
WARNING 2025-11-22 23:51:01 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 322 |
+
INFO 2025-11-22 23:51:02 ils/utils.py:46 Metal backend detected, using mps.
|
| 323 |
+
WARNING 2025-11-22 23:51:02 /policies.py:82 Device 'cuda' is not available. Switching to 'mps'.
|
| 324 |
+
INFO 2025-11-22 23:51:02 _dataset.py:316 Overriding num_inference_steps for pi05: 10
|
| 325 |
+
INFO 2025-11-22 23:52:12 ing_pi05.py:558 Enabled gradient checkpointing for PI05Pytorch model
|
| 326 |
+
WARNING 2025-11-22 23:52:13 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.bias
|
| 327 |
+
WARNING 2025-11-22 23:52:13 ng_pi05.py:1071 Vision embedding key might need handling: model.paligemma_with_expert.paligemma.model.vision_tower.vision_model.embeddings.patch_embedding.weight
|
| 328 |
+
INFO 2025-11-22 23:52:24 ng_pi05.py:1097 Using num of steps 10 as max guidance weight for RTC
|
| 329 |
+
INFO 2025-11-22 23:52:24 _dataset.py:336 RTC enabled: True
|
| 330 |
+
INFO 2025-11-22 23:52:24 _dataset.py:337 RTC debug: True
|
| 331 |
+
INFO 2025-11-22 23:52:24 _dataset.py:338 Policy config: PI05Config(n_obs_steps=1, input_features={'observation.images.image': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.images.image2': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 256, 256)), 'observation.state': PolicyFeature(type=<FeatureType.STATE: 'STATE'>, shape=(8,)), 'observation.images.empty_camera_0': PolicyFeature(type=<FeatureType.VISUAL: 'VISUAL'>, shape=(3, 224, 224))}, output_features={'action': PolicyFeature(type=<FeatureType.ACTION: 'ACTION'>, shape=(7,))}, device='mps', use_amp=False, push_to_hub=True, repo_id='pepijn223/pi05_libero_31_9_non_q', private=None, tags=None, license=None, pretrained_path=PosixPath('lerobot/pi05_libero'), paligemma_variant='gemma_2b', action_expert_variant='gemma_300m', dtype='bfloat16', chunk_size=50, n_action_steps=50, max_state_dim=32, max_action_dim=32, num_inference_steps=10, time_sampling_beta_alpha=1.5, time_sampling_beta_beta=1.0, time_sampling_scale=0.999, time_sampling_offset=0.001, min_period=0.004, max_period=4.0, rtc_config=RTCConfig(enabled=True, prefix_attention_schedule=<RTCAttentionSchedule.EXP: 'EXP'>, max_guidance_weight=10, execution_horizon=8, sigma_d=1.0, debug=True, debug_maxlen=100), image_resolution=(224, 224), empty_cameras=1, tokenizer_max_length=200, normalization_mapping={'ACTION': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'STATE': <NormalizationMode.MEAN_STD: 'MEAN_STD'>, 'VISUAL': <NormalizationMode.IDENTITY: 'IDENTITY'>}, gradient_checkpointing=True, compile_model=False, compile_mode='max-autotune', optimizer_lr=2.5e-05, optimizer_betas=(0.9, 0.95), optimizer_eps=1e-08, optimizer_weight_decay=0.01, optimizer_grad_clip_norm=1.0, scheduler_warmup_steps=1000, scheduler_decay_steps=6000, scheduler_decay_lr=2.5e-06)
|
| 332 |
+
INFO 2025-11-22 23:52:24 _dataset.py:344 ✓ policy_rtc initialized successfully
|
| 333 |
+
INFO 2025-11-22 23:52:27 _dataset.py:546 Tracked 10 steps with RTC
|
| 334 |
+
INFO 2025-11-22 23:52:27 _dataset.py:547 Generated rtc_actions shape: torch.Size([1, 50, 7])
|
| 335 |
+
ERROR 2025-11-22 23:52:27 _dataset.py:553 Error getting num_steps: 'PI05Config' object has no attribute 'num_steps'
|
| 336 |
+
WARNING 2025-11-22 23:52:27 _dataset.py:555 Using num_inference_steps: 10 instead of num_steps
|
| 337 |
+
INFO 2025-11-22 23:52:27 _dataset.py:410 Destroying policy_rtc and freeing memory...
|
| 338 |
+
INFO 2025-11-22 23:52:29 _dataset.py:430 ✓ policy_rtc destroyed and memory freed
|
| 339 |
+
INFO 2025-11-22 23:52:29 _dataset.py:561 ================================================================================
|
| 340 |
+
INFO 2025-11-22 23:52:29 _dataset.py:562 Plotting results...
|
| 341 |
+
INFO 2025-11-22 23:52:29 _dataset.py:818 ================================================================================
|
| 342 |
+
INFO 2025-11-22 23:52:29 _dataset.py:819 Plotting 10 steps
|
| 343 |
+
INFO 2025-11-22 23:52:30 _dataset.py:818 ================================================================================
|
| 344 |
+
INFO 2025-11-22 23:52:30 _dataset.py:819 Plotting 10 steps
|
| 345 |
+
INFO 2025-11-22 23:52:30 _dataset.py:905 Plotting final no-RTC x_t step as orange dashed reference
|
| 346 |
+
INFO 2025-11-22 23:52:31 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.5/denoising_xt_comparison.png
|
| 347 |
+
INFO 2025-11-22 23:52:32 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.5/denoising_vt_comparison.png
|
| 348 |
+
INFO 2025-11-22 23:52:32 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.5/denoising_correction_comparison.png
|
| 349 |
+
INFO 2025-11-22 23:52:33 _dataset.py:800 Saved figure to rtc_experiments_results/pi05_steps_10_sigma_0.5/denoising_x1t_comparison.png
|
| 350 |
+
INFO 2025-11-22 23:52:33 _dataset.py:566 ================================================================================
|
| 351 |
+
INFO 2025-11-22 23:52:33 _dataset.py:567 Plotting final actions comparison...
|
| 352 |
+
INFO 2025-11-22 23:52:34 _dataset.py:687 Saved final actions comparison to rtc_experiments_results/pi05_steps_10_sigma_0.5/final_actions_comparison.png
|
| 353 |
+
INFO 2025-11-22 23:52:34 _dataset.py:570 ================================================================================
|
| 354 |
+
INFO 2025-11-22 23:52:34 _dataset.py:571 Evaluation completed successfully
|
| 355 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 356 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 357 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 358 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 359 |
+
✓ Loaded state dict from model.safetensors
|
| 360 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 361 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 362 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 363 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 364 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 365 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 366 |
+
✓ Loaded state dict from model.safetensors
|
| 367 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 368 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 369 |
+
The PI05 model is a direct port of the OpenPI implementation.
|
| 370 |
+
This implementation follows the original OpenPI structure for compatibility.
|
| 371 |
+
Original implementation: https://github.com/Physical-Intelligence/openpi
|
| 372 |
+
Loading model from: lerobot/pi05_libero_finetuned
|
| 373 |
+
✓ Loaded state dict from model.safetensors
|
| 374 |
+
Warning: Could not remap state dict keys: Error(s) in loading state_dict for PI05Policy:
|
| 375 |
+
Missing key(s) in state_dict: "model.paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight".
|
| 376 |
+
Sat Nov 22 23:52:36 +07 2025: SUCCESS pi05 steps=10 sigma_d=0.5
|
| 377 |
+
|
| 378 |
+
Sat Nov 22 23:52:36 +07 2025: Starting pi05 steps=10 sigma_d=1.0
|
static/experiments/pi05_steps_10_sigma_0.1/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.1/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.1/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.1/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.1/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.2/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.2/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.2/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.2/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.2/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.5/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.5/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.5/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.5/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_10_sigma_0.5/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.1/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.1/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.1/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.1/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.1/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.2/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.2/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.2/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.2/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.2/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.5/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.5/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.5/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.5/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.5/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.8/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.8/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.8/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.8/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_0.8/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.0/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.0/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.0/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.0/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.0/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.2/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.2/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.2/denoising_x1t_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.2/denoising_xt_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.2/final_actions_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.5/denoising_correction_comparison.png
ADDED
|
Git LFS Details
|
static/experiments/pi05_steps_2_sigma_1.5/denoising_vt_comparison.png
ADDED
|
Git LFS Details
|