| { |
| "dataset": "pendulum_causal_qa", |
| "version": "1.0", |
| "source_dataset": "https://huggingface.co/datasets/jimchen2/pendulum-cjepa-160f", |
| "num_videos": 3000, |
| "train_videos": 2400, |
| "val_videos": 600, |
| "validation_groups": { |
| "in_distribution": 2400, |
| "val_ood_amplitude": 100, |
| "val_ood_near_conservative": 100, |
| "val_ood_initial_velocity": 100, |
| "val_ood_gravity_length": 100, |
| "val_ood_illumination": 100, |
| "val_ood_compound": 100 |
| }, |
| "num_questions_per_video": 14, |
| "num_questions": 42000, |
| "videos_generated": false, |
| "timeseries_generated": false, |
| "question_types": [ |
| "descriptive", |
| "predictive", |
| "explanatory", |
| "counterfactual" |
| ], |
| "frames_per_video": 160, |
| "fps": 12, |
| "duration_s": 13.333333333333334, |
| "size_px": [ |
| 196, |
| 196 |
| ], |
| "substeps_per_frame": 8, |
| "integrator": "RK4", |
| "ode": "theta_ddot = -(gravity/length)*sin(theta) - damping*theta_dot", |
| "train_seed": 42, |
| "validation_protocol": "Whole-video, six-stratum OOD holdout. Validation parameters use fixed low-discrepancy schedules independent of --seed; see README and index.csv.", |
| "split_unit": "whole video", |
| "qa_files": [ |
| "questions.jsonl", |
| "questions.csv" |
| ], |
| "timing": "CSV row n and decoded MP4 frame n sample time n/fps", |
| "audio": false, |
| "renderer": "upstream pendulum-cjepa-160f renderer, fixed visual rod length" |
| } |
|
|
