Research map / Video / Video generation
IRIS: A Real-World Benchmark for Inverse Recovery and Identification of Physical Dynamic Systems from Monocular Video
Rasul Khanbayov, Mohamed Rayan Barhdadi, Erchin Serpedin, HASAN KURBAN
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fix Euler integrator so acceleration term reaches physics parameters (ablation) | latent-space baseline · Delfys75 pendulum_45 | MAE of length (m): 95.07 -> 3.80 | Improved |
| fix Euler integrator gradient flow (ablation) | latent-space baseline · Delfys75 Torricelli | MAE of k: 6.07 -> 2.91; 7.67 -> 5.69 | Improved |
| fix Euler integrator gradient flow (ablation) | latent-space baseline · Delfys75 LED 2s | MAE of gamma: 2.30 -> 0.40 | Improved |
| use multi-step rollout loss instead of one-step (ablation) | +Corrected baseline · Delfys75 free fall / dropped ball | MAE of g: 1.06 -> 0.15 (free fall table) | Improved |
| use multi-step rollout loss instead of one-step (ablation) | +Corrected baseline · Delfys75 led_10s | MAE of gamma: 2.09 -> 7.46 | Hurt |
| use multi-step rollout loss instead of one-step (ablation) | +Corrected baseline · IRIS two moving pendulums | rope length MAE (m): 1.1-3.0 -> 741.1 at 20 deg | Hurt |
| use multi-step rollout loss instead of one-step (ablation) | +Corrected baseline · IRIS pendulum 90 deg | rope length MAE: +20.49 | Hurt |
| use multi-step rollout loss instead of one-step (ablation) | +Corrected baseline · IRIS rotating cone | angular stiffness MAE: -0.64, -1.21, -2.11 | Improved |
| learn coupling kappa versus kappa=0 (ablation) | hitting-cones model · IRIS hitting cones | latent reconstruction MSE: reduces MSE by 6.0-19.0% | Improved |
| train one shared multi-clip model instead of per-clip | latent-space baseline · IRIS dropping ball | MAE of g: 0.21 -> 0.05 | Improved |
| train one shared multi-clip model instead of per-clip | latent-space baseline · IRIS pendulum | MAE: 0.51 -> 22.06 | Hurt |
| sample five frames instead of one (ablation) | VLM temporal reasoning · Delfys75 | accuracy: 54% -> 73% | Improved |
Closest papers by meaning
- Learning Physics From Video: Unsupervised Physical Parameter Estimation for Continuous Dynamical Systems (CVPR 2025 · similarity 0.88)
- EMMA: Extracting Multiple physical parameters from Multimodal Data (CVPR 2026 · similarity 0.87)
- Physics from Video: Identifiability of Time-Invariant Second-Order ODEs under Minimal Trajectory Conditions (ICML 2026 · similarity 0.86)
- Apple-$\pi$: Benchmarking Thinking with Video Towards Law-Grounded Physical Intelligence (NeurIPS 2026 · similarity 0.85)
- Vid2Sim: Generalizable, Video-based Reconstruction of Appearance, Geometry and Physics for Mesh-free Simulation (CVPR 2025 · similarity 0.84)
- ImDy: Human Inverse Dynamics from Imitated Observations (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
