Add 0510 ep003 tactile_right, 0511 ep003 tactile_left/right (contact-picked windows)
Browse files- .gitattributes +21 -0
- README.md +12 -2
- e3r_diff_base.mp4 +3 -0
- e3r_diff_flow.mp4 +3 -0
- e3r_diff_fps6.mp4 +3 -0
- e3r_metrics.json +43 -0
- e3r_original.mp4 +3 -0
- e3r_recon_base.mp4 +3 -0
- e3r_recon_flow.mp4 +3 -0
- e3r_recon_fps6.mp4 +3 -0
- index.html +67 -1
- l11_diff_base.mp4 +3 -0
- l11_diff_flow.mp4 +3 -0
- l11_diff_fps6.mp4 +3 -0
- l11_metrics.json +43 -0
- l11_original.mp4 +3 -0
- l11_recon_base.mp4 +3 -0
- l11_recon_flow.mp4 +3 -0
- l11_recon_fps6.mp4 +3 -0
- r11_diff_base.mp4 +3 -0
- r11_diff_flow.mp4 +3 -0
- r11_diff_fps6.mp4 +3 -0
- r11_metrics.json +43 -0
- r11_original.mp4 +3 -0
- r11_recon_base.mp4 +3 -0
- r11_recon_flow.mp4 +3 -0
- r11_recon_fps6.mp4 +3 -0
.gitattributes
CHANGED
|
@@ -48,3 +48,24 @@ mb_original.mp4 filter=lfs diff=lfs merge=lfs -text
|
|
| 48 |
mb_recon_base.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 49 |
mb_recon_flow.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 50 |
mb_recon_fps6.mp4 filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
mb_recon_base.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 49 |
mb_recon_flow.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 50 |
mb_recon_fps6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 51 |
+
e3r_diff_base.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 52 |
+
e3r_diff_flow.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 53 |
+
e3r_diff_fps6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 54 |
+
e3r_original.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 55 |
+
e3r_recon_base.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 56 |
+
e3r_recon_flow.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 57 |
+
e3r_recon_fps6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 58 |
+
l11_diff_base.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 59 |
+
l11_diff_flow.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 60 |
+
l11_diff_fps6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 61 |
+
l11_original.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 62 |
+
l11_recon_base.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 63 |
+
l11_recon_flow.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 64 |
+
l11_recon_fps6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 65 |
+
r11_diff_base.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 66 |
+
r11_diff_flow.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 67 |
+
r11_diff_fps6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 68 |
+
r11_original.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 69 |
+
r11_recon_base.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 70 |
+
r11_recon_flow.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 71 |
+
r11_recon_fps6.mp4 filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
|
@@ -31,8 +31,18 @@ Caveat in the other direction: one training run per arm (seed 0), so a 0.28 dB g
|
|
| 31 |
|
| 32 |
## Clips
|
| 33 |
|
| 34 |
-
- **CutTofu · tactile_left** (`
|
| 35 |
-
- **
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
|
| 37 |
Per clip: `original.mp4`, `recon_{base,fps6,flow}.mp4`, `diff_{base,fps6,flow}.mp4`
|
| 38 |
(|original − recon| amplified, hot colormap), `motion_weight.mp4` (the m̂ map, where flow
|
|
|
|
| 31 |
|
| 32 |
## Clips
|
| 33 |
|
| 34 |
+
- **CutTofu · tactile_left** (`no prefix`) — 77 frames @ 256×256, 6fps, whole clip
|
| 35 |
+
- **0510 ep000 · tactile_right** (`mb_`) — 117 frames @ 256×256, 6fps, first 20s — no `*_flow.npz` sibling, so whole-frame PSNR only
|
| 36 |
+
- **0510 ep003 · tactile_right** (`e3r_`) — 117 frames @ 256×256, 6fps, first 20s — no `*_flow.npz` sibling, so whole-frame PSNR only
|
| 37 |
+
- **0511 ep003 · tactile_left** (`l11_`) — 117 frames @ 256×256, 6fps, first 20s — no `*_flow.npz` sibling, so whole-frame PSNR only
|
| 38 |
+
- **0511 ep003 · tactile_right** (`r11_`) — 117 frames @ 256×256, 6fps, 210–230s — no `*_flow.npz` sibling, so whole-frame PSNR only
|
| 39 |
+
|
| 40 |
+
Windows were picked to contain actual contact. The `motherboard_*` episodes are mostly a
|
| 41 |
+
static, defocused, no-contact gel view: a blank window still reconstructs at ~46 dB while
|
| 42 |
+
showing nothing, because there is no texture to lose. `0511_episode_003/tactile_right` is
|
| 43 |
+
blank for its first 20s, so it uses 210–230s, where the sensor is pressed against an
|
| 44 |
+
object. The clips with the most texture (CutTofu, and `0511 ep003 tactile_left`) have the
|
| 45 |
+
*lowest* base PSNR and gain the most from the LoRA — that is the signal to read.
|
| 46 |
|
| 47 |
Per clip: `original.mp4`, `recon_{base,fps6,flow}.mp4`, `diff_{base,fps6,flow}.mp4`
|
| 48 |
(|original − recon| amplified, hot colormap), `motion_weight.mp4` (the m̂ map, where flow
|
e3r_diff_base.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6e45b645e6186332fa953e8622a632bf97120b026d3e93d2db6737390e2ae4c2
|
| 3 |
+
size 3689078
|
e3r_diff_flow.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c51f1344c68dbd1200086d6f6e00f71519792854205c3eab83aeadae3fa6e25f
|
| 3 |
+
size 3472036
|
e3r_diff_fps6.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:29ae62fe8e30b47ab311b7db56c6bdbd7d745347a507f43d0c536af47a21b95a
|
| 3 |
+
size 3481883
|
e3r_metrics.json
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"input": "/group2/ct/weihanx/world_model/mode1_v2_videos/motherboard_0510_episode_003/tactile_right.mp4",
|
| 3 |
+
"n_native_frames": 600,
|
| 4 |
+
"stride": 5,
|
| 5 |
+
"seconds": 20.0,
|
| 6 |
+
"start": null,
|
| 7 |
+
"num_frames": 117,
|
| 8 |
+
"fps": 6.0,
|
| 9 |
+
"src_fps": 30.0,
|
| 10 |
+
"height": 256,
|
| 11 |
+
"width": 256,
|
| 12 |
+
"diff_amp": 20.0,
|
| 13 |
+
"flow_floor": 1.5,
|
| 14 |
+
"hi": 0.5,
|
| 15 |
+
"lo": 0.1,
|
| 16 |
+
"moving_pixel_frac": null,
|
| 17 |
+
"arms": [
|
| 18 |
+
{
|
| 19 |
+
"name": "base",
|
| 20 |
+
"adapter": null,
|
| 21 |
+
"step": null,
|
| 22 |
+
"mse": 2.884106470446568e-05,
|
| 23 |
+
"psnr": 45.399887111253115,
|
| 24 |
+
"split": null
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"name": "fps6",
|
| 28 |
+
"adapter": "./lora_out_fps6/lora_last.pt",
|
| 29 |
+
"step": 4000,
|
| 30 |
+
"mse": 2.6011546651716344e-05,
|
| 31 |
+
"psnr": 45.8483382380045,
|
| 32 |
+
"split": null
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"name": "flow",
|
| 36 |
+
"adapter": "./lora_out_flow/lora_last.pt",
|
| 37 |
+
"step": 4000,
|
| 38 |
+
"mse": 2.591218435554765e-05,
|
| 39 |
+
"psnr": 45.864959751432856,
|
| 40 |
+
"split": null
|
| 41 |
+
}
|
| 42 |
+
]
|
| 43 |
+
}
|
e3r_original.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bc4c2b0f5aa6279fafcf137b3f111b1e0fd956921e72b29965787bf90ddce9f1
|
| 3 |
+
size 254208
|
e3r_recon_base.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6e494376c7851c02c5f9eb72c296891d6bc528593cc6b0fe46d59cf74325b30e
|
| 3 |
+
size 203087
|
e3r_recon_flow.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b619caef804b2c5fc5ea9dbd9a13c093f7b136d7a230d2014f1cfff7f18da2bf
|
| 3 |
+
size 183873
|
e3r_recon_fps6.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ea05b371e6a836e6dd18ef684c4397a601d9119e80bffc173f4bc9bfdd63cc10
|
| 3 |
+
size 183010
|
index.html
CHANGED
|
@@ -73,7 +73,7 @@
|
|
| 73 |
</header>
|
| 74 |
|
| 75 |
<div class="bar">
|
| 76 |
-
<div class="tabs"><button class="tab active" data-clip="clip0">CutTofu · tactile_left</button><button class="tab" data-clip="clip1">
|
| 77 |
<div class="spacer"></div>
|
| 78 |
<button class="action" id="toggle">⏯ Play / pause visible</button>
|
| 79 |
</div>
|
|
@@ -126,6 +126,72 @@
|
|
| 126 |
these adapters is on the CutTofu tab and, at population scale, in the val table below.</p>
|
| 127 |
</section>
|
| 128 |
</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 129 |
|
| 130 |
<section>
|
| 131 |
<h2>Val set · 32 held-out clips · the population read</h2>
|
|
|
|
| 73 |
</header>
|
| 74 |
|
| 75 |
<div class="bar">
|
| 76 |
+
<div class="tabs"><button class="tab active" data-clip="clip0">CutTofu · tactile_left</button><button class="tab" data-clip="clip1">0510 ep000 · tactile_right</button><button class="tab" data-clip="clip2">0510 ep003 · tactile_right</button><button class="tab" data-clip="clip3">0511 ep003 · tactile_left</button><button class="tab" data-clip="clip4">0511 ep003 · tactile_right</button></div>
|
| 77 |
<div class="spacer"></div>
|
| 78 |
<button class="action" id="toggle">⏯ Play / pause visible</button>
|
| 79 |
</div>
|
|
|
|
| 126 |
these adapters is on the CutTofu tab and, at population scale, in the val table below.</p>
|
| 127 |
</section>
|
| 128 |
</div>
|
| 129 |
+
<div class="clip" id="clip2">
|
| 130 |
+
<div class="meta">
|
| 131 |
+
Source: <code>tactile_right.mp4</code> ·
|
| 132 |
+
Clip: <code>117 frames · 256×256 · 6fps · first 20s (stride 5 from 600 native
|
| 133 |
+
@ 30fps)</code> ·
|
| 134 |
+
Adapters: <code>lora_last.pt · r32 α32 targets=all · step 4000</code>
|
| 135 |
+
</div>
|
| 136 |
+
<section>
|
| 137 |
+
<h2>Reconstruction</h2>
|
| 138 |
+
<div class="grid"><div class="card"><div class="head"><span class="name">original</span><span class="desc">resized + subsampled input</span></div><video src="e3r_original.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">reference · 117 frames @ 256×256</div></div><div class="card"><div class="head"><span class="name">base Wan2.2</span><span class="desc">frozen VAE, no adapter</span></div><video src="e3r_recon_base.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>45.40 dB</b> · MSE <b>2.88e-05</b></div></div><div class="card"><div class="head"><span class="name">control · flow_weight 0</span><span class="desc">6fps, edge+LPIPS only</span></div><video src="e3r_recon_fps6.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>45.85 dB</b> · MSE <b>2.60e-05</b> · <span class="delta up">+0.45 dB vs base</span></div></div><div class="card"><div class="head"><span class="name">treatment · flow_weight 1</span><span class="desc">6fps, L1 weighted by motion</span></div><video src="e3r_recon_flow.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>45.86 dB</b> · MSE <b>2.59e-05</b> · <span class="delta up">+0.47 dB vs base</span></div></div></div>
|
| 139 |
+
</section>
|
| 140 |
+
<section>
|
| 141 |
+
<h2>Error maps</h2>
|
| 142 |
+
<div class="grid"><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − base Wan2.2|</span></div><video src="e3r_diff_base.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − control · flow_weight 0|</span></div><video src="e3r_diff_fps6.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − treatment · flow_weight 1|</span></div><video src="e3r_diff_flow.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div></div>
|
| 143 |
+
</section>
|
| 144 |
+
<section>
|
| 145 |
+
<h2>PSNR split by motion · this clip</h2>
|
| 146 |
+
<p class="note">No <code>*_flow.npz</code> exists beside this clip, so there is no
|
| 147 |
+
motion map to split on — only whole-frame PSNR is available here. The motion split for
|
| 148 |
+
these adapters is on the CutTofu tab and, at population scale, in the val table below.</p>
|
| 149 |
+
</section>
|
| 150 |
+
</div>
|
| 151 |
+
<div class="clip" id="clip3">
|
| 152 |
+
<div class="meta">
|
| 153 |
+
Source: <code>tactile_left.mp4</code> ·
|
| 154 |
+
Clip: <code>117 frames · 256×256 · 6fps · first 20s (stride 5 from 600 native
|
| 155 |
+
@ 30fps)</code> ·
|
| 156 |
+
Adapters: <code>lora_last.pt · r32 α32 targets=all · step 4000</code>
|
| 157 |
+
</div>
|
| 158 |
+
<section>
|
| 159 |
+
<h2>Reconstruction</h2>
|
| 160 |
+
<div class="grid"><div class="card"><div class="head"><span class="name">original</span><span class="desc">resized + subsampled input</span></div><video src="l11_original.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">reference · 117 frames @ 256×256</div></div><div class="card"><div class="head"><span class="name">base Wan2.2</span><span class="desc">frozen VAE, no adapter</span></div><video src="l11_recon_base.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>42.83 dB</b> · MSE <b>5.21e-05</b></div></div><div class="card"><div class="head"><span class="name">control · flow_weight 0</span><span class="desc">6fps, edge+LPIPS only</span></div><video src="l11_recon_fps6.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>43.55 dB</b> · MSE <b>4.41e-05</b> · <span class="delta up">+0.73 dB vs base</span></div></div><div class="card"><div class="head"><span class="name">treatment · flow_weight 1</span><span class="desc">6fps, L1 weighted by motion</span></div><video src="l11_recon_flow.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>43.59 dB</b> · MSE <b>4.37e-05</b> · <span class="delta up">+0.77 dB vs base</span></div></div></div>
|
| 161 |
+
</section>
|
| 162 |
+
<section>
|
| 163 |
+
<h2>Error maps</h2>
|
| 164 |
+
<div class="grid"><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − base Wan2.2|</span></div><video src="l11_diff_base.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − control · flow_weight 0|</span></div><video src="l11_diff_fps6.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − treatment · flow_weight 1|</span></div><video src="l11_diff_flow.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div></div>
|
| 165 |
+
</section>
|
| 166 |
+
<section>
|
| 167 |
+
<h2>PSNR split by motion · this clip</h2>
|
| 168 |
+
<p class="note">No <code>*_flow.npz</code> exists beside this clip, so there is no
|
| 169 |
+
motion map to split on — only whole-frame PSNR is available here. The motion split for
|
| 170 |
+
these adapters is on the CutTofu tab and, at population scale, in the val table below.</p>
|
| 171 |
+
</section>
|
| 172 |
+
</div>
|
| 173 |
+
<div class="clip" id="clip4">
|
| 174 |
+
<div class="meta">
|
| 175 |
+
Source: <code>tactile_right.mp4</code> ·
|
| 176 |
+
Clip: <code>117 frames · 256×256 · 6fps · 210–230s (stride 5 from 600 native
|
| 177 |
+
@ 30fps)</code> ·
|
| 178 |
+
Adapters: <code>lora_last.pt · r32 α32 targets=all · step 4000</code>
|
| 179 |
+
</div>
|
| 180 |
+
<section>
|
| 181 |
+
<h2>Reconstruction</h2>
|
| 182 |
+
<div class="grid"><div class="card"><div class="head"><span class="name">original</span><span class="desc">resized + subsampled input</span></div><video src="r11_original.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">reference · 117 frames @ 256×256</div></div><div class="card"><div class="head"><span class="name">base Wan2.2</span><span class="desc">frozen VAE, no adapter</span></div><video src="r11_recon_base.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>43.27 dB</b> · MSE <b>4.71e-05</b></div></div><div class="card"><div class="head"><span class="name">control · flow_weight 0</span><span class="desc">6fps, edge+LPIPS only</span></div><video src="r11_recon_fps6.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>43.38 dB</b> · MSE <b>4.59e-05</b> · <span class="delta up">+0.12 dB vs base</span></div></div><div class="card"><div class="head"><span class="name">treatment · flow_weight 1</span><span class="desc">6fps, L1 weighted by motion</span></div><video src="r11_recon_flow.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">PSNR <b>43.39 dB</b> · MSE <b>4.58e-05</b> · <span class="delta up">+0.13 dB vs base</span></div></div></div>
|
| 183 |
+
</section>
|
| 184 |
+
<section>
|
| 185 |
+
<h2>Error maps</h2>
|
| 186 |
+
<div class="grid"><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − base Wan2.2|</span></div><video src="r11_diff_base.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − control · flow_weight 0|</span></div><video src="r11_diff_fps6.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div><div class="card"><div class="head"><span class="name">diff ×20</span><span class="desc">|original − treatment · flow_weight 1|</span></div><video src="r11_diff_flow.mp4" muted loop playsinline controls preload="metadata"></video><div class="metric">brighter = larger error</div></div></div>
|
| 187 |
+
</section>
|
| 188 |
+
<section>
|
| 189 |
+
<h2>PSNR split by motion · this clip</h2>
|
| 190 |
+
<p class="note">No <code>*_flow.npz</code> exists beside this clip, so there is no
|
| 191 |
+
motion map to split on — only whole-frame PSNR is available here. The motion split for
|
| 192 |
+
these adapters is on the CutTofu tab and, at population scale, in the val table below.</p>
|
| 193 |
+
</section>
|
| 194 |
+
</div>
|
| 195 |
|
| 196 |
<section>
|
| 197 |
<h2>Val set · 32 held-out clips · the population read</h2>
|
l11_diff_base.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c479b732a67cd1ce055173e6b817155a189836168aad75edd754e01a1c58f8d3
|
| 3 |
+
size 3902254
|
l11_diff_flow.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:be8c111db220787d0cd0022d16ab7e20c875521a8fbeb6a67549b42e15532041
|
| 3 |
+
size 3690693
|
l11_diff_fps6.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fde41fc62d8a0aeb0e749b0fbf739bec5cc24db1f0510b244b67bd4ce824612f
|
| 3 |
+
size 3692619
|
l11_metrics.json
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"input": "/group2/ct/weihanx/world_model/mode1_v2_videos/motherboard_0511_episode_003/tactile_left.mp4",
|
| 3 |
+
"n_native_frames": 600,
|
| 4 |
+
"stride": 5,
|
| 5 |
+
"seconds": 20.0,
|
| 6 |
+
"start": null,
|
| 7 |
+
"num_frames": 117,
|
| 8 |
+
"fps": 6.0,
|
| 9 |
+
"src_fps": 30.0,
|
| 10 |
+
"height": 256,
|
| 11 |
+
"width": 256,
|
| 12 |
+
"diff_amp": 20.0,
|
| 13 |
+
"flow_floor": 1.5,
|
| 14 |
+
"hi": 0.5,
|
| 15 |
+
"lo": 0.1,
|
| 16 |
+
"moving_pixel_frac": null,
|
| 17 |
+
"arms": [
|
| 18 |
+
{
|
| 19 |
+
"name": "base",
|
| 20 |
+
"adapter": null,
|
| 21 |
+
"step": null,
|
| 22 |
+
"mse": 5.213857002672739e-05,
|
| 23 |
+
"psnr": 42.82840884138495,
|
| 24 |
+
"split": null
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"name": "fps6",
|
| 28 |
+
"adapter": "./lora_out_fps6/lora_last.pt",
|
| 29 |
+
"step": 4000,
|
| 30 |
+
"mse": 4.410687324707396e-05,
|
| 31 |
+
"psnr": 43.554937284304586,
|
| 32 |
+
"split": null
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"name": "flow",
|
| 36 |
+
"adapter": "./lora_out_flow/lora_last.pt",
|
| 37 |
+
"step": 4000,
|
| 38 |
+
"mse": 4.370337046566419e-05,
|
| 39 |
+
"psnr": 43.59485068334068,
|
| 40 |
+
"split": null
|
| 41 |
+
}
|
| 42 |
+
]
|
| 43 |
+
}
|
l11_original.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:beae3993bfedf00ca121bbf10ff14fac5da2a4d61393a9ffb26269235933b015
|
| 3 |
+
size 355054
|
l11_recon_base.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a1e1b69ef0b30ab6f0f7cc1eb29b96cedf9781af969cc3fbe4d64879d680debe
|
| 3 |
+
size 285309
|
l11_recon_flow.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:57a7790212458dee93aee5df98eaf1c622dba60cbcb9d8d6ee31138c88e48a84
|
| 3 |
+
size 257338
|
l11_recon_fps6.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:87ceba6bef0d3a866533f9245b1d621aab139c9d0fc20c5bf06e4fe6979378ec
|
| 3 |
+
size 257134
|
r11_diff_base.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3f25cfabe5f8f59fe18d18acf77ba06e28a524fd73ccb411d22a794972b131ab
|
| 3 |
+
size 3869817
|
r11_diff_flow.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7bf2bf5142f61ca9286aa9421bf534dcad38e02f0ff964d4d9eb7b04b54fcd18
|
| 3 |
+
size 3590088
|
r11_diff_fps6.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:426b9d2fe4778541dc758d2cb974eb4ba092b628d0a2cb3649a1ebf220a39643
|
| 3 |
+
size 3598632
|
r11_metrics.json
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"input": "/group2/ct/weihanx/world_model/mode1_v2_videos/motherboard_0511_episode_003/tactile_right.mp4",
|
| 3 |
+
"n_native_frames": 600,
|
| 4 |
+
"stride": 5,
|
| 5 |
+
"seconds": 20.0,
|
| 6 |
+
"start": 210.0,
|
| 7 |
+
"num_frames": 117,
|
| 8 |
+
"fps": 6.0,
|
| 9 |
+
"src_fps": 30.0,
|
| 10 |
+
"height": 256,
|
| 11 |
+
"width": 256,
|
| 12 |
+
"diff_amp": 20.0,
|
| 13 |
+
"flow_floor": 1.5,
|
| 14 |
+
"hi": 0.5,
|
| 15 |
+
"lo": 0.1,
|
| 16 |
+
"moving_pixel_frac": null,
|
| 17 |
+
"arms": [
|
| 18 |
+
{
|
| 19 |
+
"name": "base",
|
| 20 |
+
"adapter": null,
|
| 21 |
+
"step": null,
|
| 22 |
+
"mse": 4.712601366918534e-05,
|
| 23 |
+
"psnr": 43.267392951161455,
|
| 24 |
+
"split": null
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"name": "fps6",
|
| 28 |
+
"adapter": "./lora_out_fps6/lora_last.pt",
|
| 29 |
+
"step": 4000,
|
| 30 |
+
"mse": 4.5883636630605906e-05,
|
| 31 |
+
"psnr": 43.38342168233397,
|
| 32 |
+
"split": null
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"name": "flow",
|
| 36 |
+
"adapter": "./lora_out_flow/lora_last.pt",
|
| 37 |
+
"step": 4000,
|
| 38 |
+
"mse": 4.577119761961512e-05,
|
| 39 |
+
"psnr": 43.39407723954923,
|
| 40 |
+
"split": null
|
| 41 |
+
}
|
| 42 |
+
]
|
| 43 |
+
}
|
r11_original.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c1c050e2fa33a552f6560e34dff7e5af321cdb83ca4eac0d52614b2b6572207a
|
| 3 |
+
size 340927
|
r11_recon_base.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3451028a7aa587532e8598a2ef21728e078228460d575f10837162eb9ce623e8
|
| 3 |
+
size 310913
|
r11_recon_flow.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5a979c562c370690ce74b4c871823e57886b842741b7e03648709c026b3c0fbc
|
| 3 |
+
size 259467
|
r11_recon_fps6.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d0c27d808a9a27a5b214b88094402571fa3ff324b7e3105e0bebc3ec5730def2
|
| 3 |
+
size 259086
|