Research map / Video / Video segmentation and tracking
Physics-as-Inverse-Graphics: Unsupervised Physical Parameter Estimation from Video
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add joint training with prediction and autoencoder losses instead of using prediction loss alone (ablation) | Physics-as-Inverse-Graphics · 3-balls gravity | test loss: 31.4 -> 1.39 | Helped |
| Add joint training with prediction and autoencoder losses instead of training the vision and physics modules with separate gradients (ablation) | Physics-as-Inverse-Graphics · 3-balls gravity | test loss: 28.1 -> 1.39 | Helped |
| Use a coordinate-consistent graphics decoder instead of a black-box MLP decoder (ablation) | Physics-as-Inverse-Graphics · 3-balls gravity | test loss: 30.9 -> 1.39 | Helped |
| Add the autoencoder loss to joint prediction-loss training (ablation) | Physics-as-Inverse-Graphics · 3-balls gravity | reconstruction loss: 20.5 -> 0.63 | Helped |
| Train vision and physics modules jointly rather than with separate gradients (ablation) | Physics-as-Inverse-Graphics · 3-balls gravity | reconstruction loss: 0.22 -> 0.63 | Hurt |
| Use a coordinate-consistent graphics decoder instead of a black-box MLP decoder (ablation) | Physics-as-Inverse-Graphics · 3-balls gravity | reconstruction loss: 2.87 -> 0.63 | Helped |
| Use explicit physical equations for dynamics rather than learned black-box dynamics | Physics-as-Inverse-Graphics · Balls and digits prediction extrapolation | SSI | Helped |
| Use a graphics renderer rather than Deconv or Spatial Broadcast black-box decoders (ablation) | Physics-as-Inverse-Graphics · Unseen bottom-half image regions | object integrity in extrapolated frames | Helped |
| Use too few or too many object slots instead of the correct number (ablation) | Physics-as-Inverse-Graphics · 3-balls gravity | predictive performance | Hurt |
Closest papers by meaning
- Learning Physics From Video: Unsupervised Physical Parameter Estimation for Continuous Dynamical Systems (CVPR 2025 · similarity 0.88)
- Seeing the Wind from a Falling Leaf (NeurIPS 2025 · similarity 0.87)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.86)
- MonoPhysics: Estimating Geometry, Appearance, and Physical Parameters from Monocular Videos (NeurIPS 2026 · similarity 0.86)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.86)
- Latent Intuitive Physics: Learning to Transfer Hidden Physics from A 3D Video (ICLR 2024 · similarity 0.85)
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.