video video 13.3 13.3 | label class label 0
classes |
|---|---|
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null | |
null |
Pendulum C-JEPA 3K + CLEVRER-style Questions (bob higher above the shadow)
3,000 silent H.264 MP4s of a nonlinear physical pendulum (160 frames @ 12 fps = 13.33 s, 196×196 px), with frame-aligned CSV labels and ~24,000 CLEVRER-style questions (descriptive / explanatory / predictive / counterfactual) whose answers are all derived from the ground-truth simulation.
This is a sibling of jimchen2/jimchen2-pendulum-output-11
(jimchen2/pendulum-cjepa-160f recipe).
Content, physics, split design, seeds, and question templates are identical — the only
change is the renderer geometry: the drawn rod is shorter and the ground line lower,
so the bob sits noticeably higher above its shadow.
| constant | -output-11 |
this repo |
|---|---|---|
DRAW_LENGTH (drawn rod, world units) |
8.0 |
6.5 |
GROUND_Y (ground / shadow line) |
-0.5 |
-1.6 |
| bob height at bottom (θ=0) | 2.5 |
4.0 |
| gap bob-bottom → shadow at the bottom | ~1.5 units | ~4.1 units |
Physical simulation (LENGTH_M=1.0, gravity, damping models, driving), frame count,
FPS, image size, CSV schema, and question templates are unchanged. CSV geometry
columns (bob_x/y_world, shadow_left/right_world, …) are recomputed from the new
constants, so labels stay frame-aligned with the rendered videos, and the
shadow-related questions were re-derived with the same geometry.
| split | videos | physics |
|---|---|---|
| train | 2,400 | in-distribution (identical recipe to the original dataset) |
| val | 600 (6 regimes × 100) | OOD — see regimes below |
OOD validation regimes
| regime | what changes vs train | why it is a causal test |
|---|---|---|
val_high_damp |
γ ∈ U(0.55, 0.90) s⁻¹ (train: 0.15–0.50) | envelope time-constant τ=2/γ as low as 2.2 s, shorter than anything seen |
val_low_damp |
γ ∈ U(0, 0.08) s⁻¹ | near-conservative dynamics; amplitude/energy almost constant |
val_large_angle |
θ₀ = ±(45–80)° (train: ±15–40°) | strong nonlinearity: period elongated up to ~30%, visibly non-sinusoidal |
val_gravity |
g ∈ {1.62 Moon, 3.71 Mars, 24.79 Jupiter} m/s² | period shifts by up to 2.5× both directions |
val_quad_damping |
drag = −β·ω·|ω|, β ∈ U(0.10, 0.35); γ = 0 | new force law: amplitude envelope decays ~linearly, not exponentially |
val_driven |
periodic torque A·cos(2πf t+φ), A ∈ U(0.35, 1.0) rad/s², f ∈ U(0.35, 0.65) Hz | new causal mechanism: beats / resonant growth around the ~0.5 Hz natural frequency; energy no longer decays monotonically |
Config rejection-sampling guarantees every simulated frame stays inside the scene and |θ| ≤ 85° (the driven regime is the main source of rejections, near resonance).
Question set (CLEVRER taxonomy)
~8 questions per video, ~24,000 total. Format mirrors CLEVRER:
{
"info": {...},
"scene_annotation": [
{"video_id": "video_00000",
"video_filename": "videos/train/video_00000.mp4",
"split": "train", "regime": "train",
"questions": [
{"question_id": 0, "question_type": "descriptive",
"question": "...", "answer": "left"},
{"question_id": 1, "question_type": "counterfactual",
"question": "...", "choices": ["...", "..."], "answer": 0}
]}
]
}
- descriptive — observable facts: release side, bottom-crossing count, end-of-clip motion, damping envelope phrase, light drift, energy shape, fastest point, shadow direction. Mostly free-form string answers; three templates are convenience multiple-choice.
- explanatory (4-way MC) — what is responsible for what you see: amplitude decay (friction), stopping at the top (velocity through zero), fastest at the bottom (energy exchange), shadow sweeping (bob blocks light), never returning to release height (dissipated energy), beats in driven clips (rhythmic external push), large-angle period elongation (sin θ vs θ).
- predictive (4-way MC) — next peak higher/lower/same, motion right after the clip ends, bottom crossing within the next second, long-term fate (settle / keep swinging / driven pattern), future shadow motion. All computed from a hidden 28-second ground-truth continuation of the simulation (videos show only the first 13.33 s), so these have real answers.
- counterfactual (4-way MC) — interventions: remove damping (returns to release height), add damping to an undamped clip, double gravity (faster), double rod length (slower), double bob mass (no change — mass independence), move/brighten the light (no effect on motion — the illumination is an independent nuisance factor by construction), steeper release (period slightly longer — amplitude coupling), double damping (dies out ~twice as fast), turn the drive off (settles), vacuum removes quadratic drag, Earth ↔ other-planet gravity.
Each question carries a concept tag (e.g. damping envelope,
interventional invariance, g dependence of period) so you can score accuracy
per causal concept and per OOD regime.
Layout
.
├── videos/train/video_00000.mp4 ... 2400 clips
├── videos/val/video_10000.mp4 ... 600 clips (val ids offset by 10000)
├── timeseries/video_XXXXX.csv 3000 frame-aligned label files
├── questions/questions_train.json 19,200 questions
├── questions/questions_val.json 4,800 questions
├── questions/taxonomy.json question-type registry
├── wds/train/pendulum-train-000000.tar ... 19 WebDataset shards (128 videos each, last has 96)
├── wds/val/pendulum-val-000000.tar ... 5 WebDataset shards (128 videos each, last has 88)
├── index.csv per-video params, split, regime
├── metadata.json
├── sample_sheet.png
└── scripts/
├── pendulum_gen.py physics + renderer + regime samplers
├── questions.py CLEVRER-style question generator
├── run_generate.py parallel generator (resume-safe)
├── verify_dataset.py integrity checks
├── audit_answers.py re-derive & audit question answers
├── make_preview.py build preview.html
├── prepare_cjepa_root.sh recreate cjepa_data_root/ for extract_four.py
└── upload_to_hf.py push this folder to the Hub
WebDataset shards
wds/ holds WebDataset tars built from the videos with
webdataset ShardWriter(..., maxcount=128);
each sample is <stem>.video.mp4 + <stem>.meta.json
({"split": ..., "filename": ...}), keyed by <stem>. The builder is the same
snippet shown in the source dataset (from webdataset import ShardWriter), pointed
at this repo's videos/ folder. Shard files are tracked with Git LFS.
Frame n ↔ CSV row frame=n ↔ time_s = n/12. Splits are whole-video.
Physics
theta'' = -(g/L)·sin(theta) - gamma·theta' - beta·theta'·|theta'| + A·cos(2π f_d t + φ)
RK4 at fps × 8 = 96 Hz. Train: g=9.81, L=1.0, γ∈U(0.15,0.50), β=0, A=0, θ₀=±U(15,40)°, ω₀=U(−0.2,0.2), light mean U(70,110)°, light amp U(0,15)°, light freq U(0.04,0.12) Hz, random phase — identical to the original dataset. Illumination is independent of the dynamics (interventional-invariance property used by the counterfactual light questions).
energy_j_per_kg, theta_rad, omega_rad_s, alpha_rad_s2, bob/light/shadow
geometry are all in the CSVs, so numeric extrapolation error at frames 128→159
(+2.67 s ≈ 1.3 oscillations) can also be scored exactly, as in the original.
Reproducing / extending
python scripts/run_generate.py --root dataset --n-train 2400 --val-per-regime 100 --seed 20261014
python scripts/verify_dataset.py dataset
Generation is seed-deterministic (per-video SeedSequence([seed, index])), resume-safe
(re-running skips finished videos), and was produced with the exact scripts in scripts/
(geometry constants DRAW_LENGTH=6.5, GROUND_Y=-1.6).
License
MIT. Renderer/physics adapted from jimchen2/pendulum-cjepa-160f (whose own code
credits Huawei Technologies Co., Ltd., also MIT).
- Downloads last month
- 93