helper2424 commited on
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
Files changed (50) hide show
  1. .gitattributes +1 -0
  2. index.html +372 -374
  3. static/experiments/experiment_log.txt +378 -0
  4. static/experiments/pi05_steps_10_sigma_0.1/denoising_correction_comparison.png +3 -0
  5. static/experiments/pi05_steps_10_sigma_0.1/denoising_vt_comparison.png +3 -0
  6. static/experiments/pi05_steps_10_sigma_0.1/denoising_x1t_comparison.png +3 -0
  7. static/experiments/pi05_steps_10_sigma_0.1/denoising_xt_comparison.png +3 -0
  8. static/experiments/pi05_steps_10_sigma_0.1/final_actions_comparison.png +3 -0
  9. static/experiments/pi05_steps_10_sigma_0.2/denoising_correction_comparison.png +3 -0
  10. static/experiments/pi05_steps_10_sigma_0.2/denoising_vt_comparison.png +3 -0
  11. static/experiments/pi05_steps_10_sigma_0.2/denoising_x1t_comparison.png +3 -0
  12. static/experiments/pi05_steps_10_sigma_0.2/denoising_xt_comparison.png +3 -0
  13. static/experiments/pi05_steps_10_sigma_0.2/final_actions_comparison.png +3 -0
  14. static/experiments/pi05_steps_10_sigma_0.5/denoising_correction_comparison.png +3 -0
  15. static/experiments/pi05_steps_10_sigma_0.5/denoising_vt_comparison.png +3 -0
  16. static/experiments/pi05_steps_10_sigma_0.5/denoising_x1t_comparison.png +3 -0
  17. static/experiments/pi05_steps_10_sigma_0.5/denoising_xt_comparison.png +3 -0
  18. static/experiments/pi05_steps_10_sigma_0.5/final_actions_comparison.png +3 -0
  19. static/experiments/pi05_steps_2_sigma_0.1/denoising_correction_comparison.png +3 -0
  20. static/experiments/pi05_steps_2_sigma_0.1/denoising_vt_comparison.png +3 -0
  21. static/experiments/pi05_steps_2_sigma_0.1/denoising_x1t_comparison.png +3 -0
  22. static/experiments/pi05_steps_2_sigma_0.1/denoising_xt_comparison.png +3 -0
  23. static/experiments/pi05_steps_2_sigma_0.1/final_actions_comparison.png +3 -0
  24. static/experiments/pi05_steps_2_sigma_0.2/denoising_correction_comparison.png +3 -0
  25. static/experiments/pi05_steps_2_sigma_0.2/denoising_vt_comparison.png +3 -0
  26. static/experiments/pi05_steps_2_sigma_0.2/denoising_x1t_comparison.png +3 -0
  27. static/experiments/pi05_steps_2_sigma_0.2/denoising_xt_comparison.png +3 -0
  28. static/experiments/pi05_steps_2_sigma_0.2/final_actions_comparison.png +3 -0
  29. static/experiments/pi05_steps_2_sigma_0.5/denoising_correction_comparison.png +3 -0
  30. static/experiments/pi05_steps_2_sigma_0.5/denoising_vt_comparison.png +3 -0
  31. static/experiments/pi05_steps_2_sigma_0.5/denoising_x1t_comparison.png +3 -0
  32. static/experiments/pi05_steps_2_sigma_0.5/denoising_xt_comparison.png +3 -0
  33. static/experiments/pi05_steps_2_sigma_0.5/final_actions_comparison.png +3 -0
  34. static/experiments/pi05_steps_2_sigma_0.8/denoising_correction_comparison.png +3 -0
  35. static/experiments/pi05_steps_2_sigma_0.8/denoising_vt_comparison.png +3 -0
  36. static/experiments/pi05_steps_2_sigma_0.8/denoising_x1t_comparison.png +3 -0
  37. static/experiments/pi05_steps_2_sigma_0.8/denoising_xt_comparison.png +3 -0
  38. static/experiments/pi05_steps_2_sigma_0.8/final_actions_comparison.png +3 -0
  39. static/experiments/pi05_steps_2_sigma_1.0/denoising_correction_comparison.png +3 -0
  40. static/experiments/pi05_steps_2_sigma_1.0/denoising_vt_comparison.png +3 -0
  41. static/experiments/pi05_steps_2_sigma_1.0/denoising_x1t_comparison.png +3 -0
  42. static/experiments/pi05_steps_2_sigma_1.0/denoising_xt_comparison.png +3 -0
  43. static/experiments/pi05_steps_2_sigma_1.0/final_actions_comparison.png +3 -0
  44. static/experiments/pi05_steps_2_sigma_1.2/denoising_correction_comparison.png +3 -0
  45. static/experiments/pi05_steps_2_sigma_1.2/denoising_vt_comparison.png +3 -0
  46. static/experiments/pi05_steps_2_sigma_1.2/denoising_x1t_comparison.png +3 -0
  47. static/experiments/pi05_steps_2_sigma_1.2/denoising_xt_comparison.png +3 -0
  48. static/experiments/pi05_steps_2_sigma_1.2/final_actions_comparison.png +3 -0
  49. static/experiments/pi05_steps_2_sigma_1.5/denoising_correction_comparison.png +3 -0
  50. 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
- content="Deformable Neural Radiance Fields creates free-viewpoint portraits (nerfies) from casually captured videos.">
7
- <meta name="keywords" content="Nerfies, D-NeRF, NeRF">
8
  <meta name="viewport" content="width=device-width, initial-scale=1">
9
- <title>Nerfies: Deformable Neural Radiance Fields</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
- <script src="./static/js/bulma-carousel.min.js"></script>
26
- <script src="./static/js/bulma-slider.min.js"></script>
27
- <script src="./static/js/index.js"></script>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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">Nerfies: Deformable Neural Radiance Fields</h1>
37
- <div class="is-size-5 publication-authors">
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"><sup>1</sup>University of Washington,</span>
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="hero teaser">
124
  <div class="container is-max-desktop">
125
- <div class="hero-body">
126
- <video id="teaser" autoplay muted loop playsinline height="100%">
127
- <source src="./static/videos/teaser.mp4"
128
- type="video/mp4">
129
- </video>
130
- <h2 class="subtitle has-text-centered">
131
- <span class="dnerf">Nerfies</span> turns selfie videos from your phone into
132
- free-viewpoint
133
- portraits.
134
- </h2>
 
 
 
 
135
  </div>
136
- </div>
137
- </section>
138
 
139
-
140
- <section class="hero is-light is-small">
141
- <div class="hero-body">
142
- <div class="container">
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
- <section class="section">
199
- <div class="container is-max-desktop">
200
- <!-- Abstract. -->
201
- <div class="columns is-centered has-text-centered">
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
- </div>
235
- <!--/ Abstract. -->
236
-
237
- <!-- Paper video. -->
238
- <div class="columns is-centered has-text-centered">
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&amp;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
- <section class="section">
253
- <div class="container is-max-desktop">
 
254
 
255
- <div class="columns is-centered">
256
-
257
- <!-- Visual Effects. -->
258
- <div class="column">
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
- </div>
 
 
 
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
- </div>
386
- </section>
 
 
 
387
 
 
 
 
 
 
388
 
389
- <section class="section" id="BibTeX">
390
- <div class="container is-max-desktop content">
391
- <h2 class="title">BibTeX</h2>
392
- <pre><code>@article{park2021nerfies,
393
- author = {Park, Keunhong and Sinha, Utkarsh and Barron, Jonathan T. and Bouaziz, Sofien and Goldman, Dan B and Seitz, Steven M. and Martin-Brualla, Ricardo},
394
- title = {Nerfies: Deformable Neural Radiance Fields},
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
- <a class="icon-link" target="_blank"
406
- href="./static/videos/nerfies_paper.pdf">
407
- <i class="fas fa-file-pdf"></i>
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">&times;</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

  • SHA256: 8586b49d1e7489b52bc60b825b8ad3b51afa5f248b2f0ad6a32ed50803f2eabe
  • Pointer size: 131 Bytes
  • Size of remote file: 251 kB
static/experiments/pi05_steps_10_sigma_0.1/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 058a3c8c7eea73ded9c8ff872db625a5596dd8004487c9fdbb45f435563ee3f5
  • Pointer size: 131 Bytes
  • Size of remote file: 814 kB
static/experiments/pi05_steps_10_sigma_0.1/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: a32d06abd1b9be9b62a8ff7ac5ab030c5256e569307b826f36628b672c114beb
  • Pointer size: 131 Bytes
  • Size of remote file: 487 kB
static/experiments/pi05_steps_10_sigma_0.1/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 5ba5494700940f8d17823316e66ff2a574f703f3275a44af5b81648a134af8e1
  • Pointer size: 132 Bytes
  • Size of remote file: 1.37 MB
static/experiments/pi05_steps_10_sigma_0.1/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: 0e098a96e5e89b2088b31e5a7fb0c81244b31205dfe5eb3bd02c2dac6ed81aa1
  • Pointer size: 131 Bytes
  • Size of remote file: 332 kB
static/experiments/pi05_steps_10_sigma_0.2/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 8586b49d1e7489b52bc60b825b8ad3b51afa5f248b2f0ad6a32ed50803f2eabe
  • Pointer size: 131 Bytes
  • Size of remote file: 251 kB
static/experiments/pi05_steps_10_sigma_0.2/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 058a3c8c7eea73ded9c8ff872db625a5596dd8004487c9fdbb45f435563ee3f5
  • Pointer size: 131 Bytes
  • Size of remote file: 814 kB
static/experiments/pi05_steps_10_sigma_0.2/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: a32d06abd1b9be9b62a8ff7ac5ab030c5256e569307b826f36628b672c114beb
  • Pointer size: 131 Bytes
  • Size of remote file: 487 kB
static/experiments/pi05_steps_10_sigma_0.2/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 5ba5494700940f8d17823316e66ff2a574f703f3275a44af5b81648a134af8e1
  • Pointer size: 132 Bytes
  • Size of remote file: 1.37 MB
static/experiments/pi05_steps_10_sigma_0.2/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: 0e098a96e5e89b2088b31e5a7fb0c81244b31205dfe5eb3bd02c2dac6ed81aa1
  • Pointer size: 131 Bytes
  • Size of remote file: 332 kB
static/experiments/pi05_steps_10_sigma_0.5/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 8586b49d1e7489b52bc60b825b8ad3b51afa5f248b2f0ad6a32ed50803f2eabe
  • Pointer size: 131 Bytes
  • Size of remote file: 251 kB
static/experiments/pi05_steps_10_sigma_0.5/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 058a3c8c7eea73ded9c8ff872db625a5596dd8004487c9fdbb45f435563ee3f5
  • Pointer size: 131 Bytes
  • Size of remote file: 814 kB
static/experiments/pi05_steps_10_sigma_0.5/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: a32d06abd1b9be9b62a8ff7ac5ab030c5256e569307b826f36628b672c114beb
  • Pointer size: 131 Bytes
  • Size of remote file: 487 kB
static/experiments/pi05_steps_10_sigma_0.5/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 5ba5494700940f8d17823316e66ff2a574f703f3275a44af5b81648a134af8e1
  • Pointer size: 132 Bytes
  • Size of remote file: 1.37 MB
static/experiments/pi05_steps_10_sigma_0.5/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: 0e098a96e5e89b2088b31e5a7fb0c81244b31205dfe5eb3bd02c2dac6ed81aa1
  • Pointer size: 131 Bytes
  • Size of remote file: 332 kB
static/experiments/pi05_steps_2_sigma_0.1/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 61353ace1a51150457d3d3bbc4c04a0cdf49c9367eef7a20294e996e3d64b011
  • Pointer size: 131 Bytes
  • Size of remote file: 181 kB
static/experiments/pi05_steps_2_sigma_0.1/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 27559cdb443eb298c62eaae206cc00809e45dc84da87521cdafa14cedac1fa40
  • Pointer size: 131 Bytes
  • Size of remote file: 610 kB
static/experiments/pi05_steps_2_sigma_0.1/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: b996bf912d882a2cad2e7038de6f61e71b277099e9779ded0fd3149b092bd0bd
  • Pointer size: 131 Bytes
  • Size of remote file: 324 kB
static/experiments/pi05_steps_2_sigma_0.1/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 4d2d99962dcaabab31da4d12743423bd0b5090a4d8934015503e82464b497feb
  • Pointer size: 131 Bytes
  • Size of remote file: 677 kB
static/experiments/pi05_steps_2_sigma_0.1/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: bc1cd66972bda02b3e78d4dbc1da02fa5d402d09310a5dce1d98f17e635ab194
  • Pointer size: 131 Bytes
  • Size of remote file: 315 kB
static/experiments/pi05_steps_2_sigma_0.2/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 61353ace1a51150457d3d3bbc4c04a0cdf49c9367eef7a20294e996e3d64b011
  • Pointer size: 131 Bytes
  • Size of remote file: 181 kB
static/experiments/pi05_steps_2_sigma_0.2/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 27559cdb443eb298c62eaae206cc00809e45dc84da87521cdafa14cedac1fa40
  • Pointer size: 131 Bytes
  • Size of remote file: 610 kB
static/experiments/pi05_steps_2_sigma_0.2/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: b996bf912d882a2cad2e7038de6f61e71b277099e9779ded0fd3149b092bd0bd
  • Pointer size: 131 Bytes
  • Size of remote file: 324 kB
static/experiments/pi05_steps_2_sigma_0.2/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 4d2d99962dcaabab31da4d12743423bd0b5090a4d8934015503e82464b497feb
  • Pointer size: 131 Bytes
  • Size of remote file: 677 kB
static/experiments/pi05_steps_2_sigma_0.2/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: bc1cd66972bda02b3e78d4dbc1da02fa5d402d09310a5dce1d98f17e635ab194
  • Pointer size: 131 Bytes
  • Size of remote file: 315 kB
static/experiments/pi05_steps_2_sigma_0.5/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 61353ace1a51150457d3d3bbc4c04a0cdf49c9367eef7a20294e996e3d64b011
  • Pointer size: 131 Bytes
  • Size of remote file: 181 kB
static/experiments/pi05_steps_2_sigma_0.5/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 27559cdb443eb298c62eaae206cc00809e45dc84da87521cdafa14cedac1fa40
  • Pointer size: 131 Bytes
  • Size of remote file: 610 kB
static/experiments/pi05_steps_2_sigma_0.5/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: b996bf912d882a2cad2e7038de6f61e71b277099e9779ded0fd3149b092bd0bd
  • Pointer size: 131 Bytes
  • Size of remote file: 324 kB
static/experiments/pi05_steps_2_sigma_0.5/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 4d2d99962dcaabab31da4d12743423bd0b5090a4d8934015503e82464b497feb
  • Pointer size: 131 Bytes
  • Size of remote file: 677 kB
static/experiments/pi05_steps_2_sigma_0.5/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: bc1cd66972bda02b3e78d4dbc1da02fa5d402d09310a5dce1d98f17e635ab194
  • Pointer size: 131 Bytes
  • Size of remote file: 315 kB
static/experiments/pi05_steps_2_sigma_0.8/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 61353ace1a51150457d3d3bbc4c04a0cdf49c9367eef7a20294e996e3d64b011
  • Pointer size: 131 Bytes
  • Size of remote file: 181 kB
static/experiments/pi05_steps_2_sigma_0.8/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 27559cdb443eb298c62eaae206cc00809e45dc84da87521cdafa14cedac1fa40
  • Pointer size: 131 Bytes
  • Size of remote file: 610 kB
static/experiments/pi05_steps_2_sigma_0.8/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: b996bf912d882a2cad2e7038de6f61e71b277099e9779ded0fd3149b092bd0bd
  • Pointer size: 131 Bytes
  • Size of remote file: 324 kB
static/experiments/pi05_steps_2_sigma_0.8/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 4d2d99962dcaabab31da4d12743423bd0b5090a4d8934015503e82464b497feb
  • Pointer size: 131 Bytes
  • Size of remote file: 677 kB
static/experiments/pi05_steps_2_sigma_0.8/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: bc1cd66972bda02b3e78d4dbc1da02fa5d402d09310a5dce1d98f17e635ab194
  • Pointer size: 131 Bytes
  • Size of remote file: 315 kB
static/experiments/pi05_steps_2_sigma_1.0/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 61353ace1a51150457d3d3bbc4c04a0cdf49c9367eef7a20294e996e3d64b011
  • Pointer size: 131 Bytes
  • Size of remote file: 181 kB
static/experiments/pi05_steps_2_sigma_1.0/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 27559cdb443eb298c62eaae206cc00809e45dc84da87521cdafa14cedac1fa40
  • Pointer size: 131 Bytes
  • Size of remote file: 610 kB
static/experiments/pi05_steps_2_sigma_1.0/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: b996bf912d882a2cad2e7038de6f61e71b277099e9779ded0fd3149b092bd0bd
  • Pointer size: 131 Bytes
  • Size of remote file: 324 kB
static/experiments/pi05_steps_2_sigma_1.0/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 4d2d99962dcaabab31da4d12743423bd0b5090a4d8934015503e82464b497feb
  • Pointer size: 131 Bytes
  • Size of remote file: 677 kB
static/experiments/pi05_steps_2_sigma_1.0/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: bc1cd66972bda02b3e78d4dbc1da02fa5d402d09310a5dce1d98f17e635ab194
  • Pointer size: 131 Bytes
  • Size of remote file: 315 kB
static/experiments/pi05_steps_2_sigma_1.2/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 61353ace1a51150457d3d3bbc4c04a0cdf49c9367eef7a20294e996e3d64b011
  • Pointer size: 131 Bytes
  • Size of remote file: 181 kB
static/experiments/pi05_steps_2_sigma_1.2/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 27559cdb443eb298c62eaae206cc00809e45dc84da87521cdafa14cedac1fa40
  • Pointer size: 131 Bytes
  • Size of remote file: 610 kB
static/experiments/pi05_steps_2_sigma_1.2/denoising_x1t_comparison.png ADDED

Git LFS Details

  • SHA256: b996bf912d882a2cad2e7038de6f61e71b277099e9779ded0fd3149b092bd0bd
  • Pointer size: 131 Bytes
  • Size of remote file: 324 kB
static/experiments/pi05_steps_2_sigma_1.2/denoising_xt_comparison.png ADDED

Git LFS Details

  • SHA256: 4d2d99962dcaabab31da4d12743423bd0b5090a4d8934015503e82464b497feb
  • Pointer size: 131 Bytes
  • Size of remote file: 677 kB
static/experiments/pi05_steps_2_sigma_1.2/final_actions_comparison.png ADDED

Git LFS Details

  • SHA256: bc1cd66972bda02b3e78d4dbc1da02fa5d402d09310a5dce1d98f17e635ab194
  • Pointer size: 131 Bytes
  • Size of remote file: 315 kB
static/experiments/pi05_steps_2_sigma_1.5/denoising_correction_comparison.png ADDED

Git LFS Details

  • SHA256: 61353ace1a51150457d3d3bbc4c04a0cdf49c9367eef7a20294e996e3d64b011
  • Pointer size: 131 Bytes
  • Size of remote file: 181 kB
static/experiments/pi05_steps_2_sigma_1.5/denoising_vt_comparison.png ADDED

Git LFS Details

  • SHA256: 27559cdb443eb298c62eaae206cc00809e45dc84da87521cdafa14cedac1fa40
  • Pointer size: 131 Bytes
  • Size of remote file: 610 kB