PhysInOne: Visual Physics Learning and Reasoning in One Suite

vLAR Group | The Hong Kong Polytechnic University | Syai Singapore | Meta

CVPR 2026

🧭 Navigation

📌 Summary
🚀 Release Timetable
📦 Repositories & Downloads
📊 Data Splits
🧱 3D Assets
🛠️ Data Processing
🏆 Leaderboard Evaluation Data
🎞️ Rendered Data
   1. Download Scripts
   2. Install Dependencies
   3. Filter Cases
      3.1 Split
      3.2 Split + Activity
      3.3 Phenomenon
      3.4 Random Sampling
   4. Download Cases
📦 Dataset Structure
📝 Annotation Details
   • Depth
   • Segmentation
   • Captions
   • Trajectories
      ◦ JSON file structure
      ◦ Actor-level pose storage
      ◦ Pose entry format
   • Cameras
      ◦ Camera JSON structure
      ◦ Dynamic camera behavior
   • Point Clouds
🧩 Data Fields
   • Physical Phenomena
📜 License
📚 Citation
📮 Contact

📌 Summary

PhysInOne is a large-scale synthetic dataset for visual physics learning and reasoning, containing 153,810 dynamic 3D scenes and 2 million annotated videos across 71 physical phenomena in mechanics, optics, fluid dynamics, and magnetism. Each scene features complex multi-object and multi-physics interactions with rich annotations, including RGB videos, depth maps, object masks, 3D trajectories, camera poses, object meshes, material properties, and textual descriptions. The dataset supports research on physics-aware video generation, future frame prediction, physical property estimation, motion transfer, physical reasoning, and world models.

153,810
Dynamic 3D scenes

2M
Annotated videos

71
Physical phenomena

4
Physics domains

🚀 Release Timetable

Component	Progress	Status	Notes
Rendered Data - Train	██████████ 100%(122988/122988)	Released	Last updated: Aug 21
Rendered Data - Val	░░░░░░░░░░ 1%(103/15411)	In progress
3D Assets	██████████ 100%	Released	14,847 Val scenes, 120,660 Train scenes, and Train/Val resource libraries
Leaderboard Evaluation Inputs	██████████ 100%	Released	All public task inputs released; ground truth excluded
Baseline code	███████▌░░ 75%	In progress	Six long-term future prediction baselines released; updated Oct 8
Data processing	██████████ 100%	Released	Camera placement, UE rendering, EXR conversion, and point-cloud generation

💻 Baselines

Six baseline implementations for Long-term Future Frame Prediction are now available in the public GitHub repository.

• Models: FreeGave, Def3G, ExtDM, MAGI-1, TRACE, and TiNeuVox

Each package contains cleaned source code, English setup and usage instructions, a directory-layout guide, the 103-scene evaluation manifest, and checkpoint-placement instructions. Training instructions are included where supported; MAGI-1 is provided as an inference-only integration. Model weights and benchmark ground truth are not bundled with the source release.

📦 Dataset Repositories & Downloads

Due to the large scale of PhysInOne, the rendered data and annotations are split across 16 Hugging Face repositories. Each entry shows the shard size, release status, live all-time downloads, live downloads in the last 30 days, and its repository link.

Combined snapshot (Oct 1, 2026): P01–P16 have 1,233,150 all-time downloads and 408,806 downloads in the last 30 days. Including the main repository, the per-repository sums are 1,280,414 and 442,702.

Main repositoryMainOpen ↗

PhysInOneP01Train · 01/16CompleteOpen ↗

4.52 TB

PhysInOneP02Train · 02/16CompleteOpen ↗

7.09 TB

PhysInOneP03Train · 03/16CompleteOpen ↗

7.59 TB

PhysInOneP04Train · 04/16CompleteOpen ↗

7.58 TB

PhysInOneP05Train · 05/16CompleteOpen ↗

7.16 TB

PhysInOneP06Train · 06/16CompleteOpen ↗

7.19 TB

PhysInOneP07Train · 07/16CompleteOpen ↗

7.21 TB

PhysInOneP08Train · 08/16CompleteOpen ↗

7.21 TB

PhysInOneP09Train · 09/16CompleteOpen ↗

7.20 TB

PhysInOneP10Train · 10/16CompleteOpen ↗

7.25 TB

PhysInOneP11Train · 11/16CompleteOpen ↗

7.48 TB

PhysInOneP12Train · 12/16CompleteOpen ↗

6.68 TB

PhysInOneP13Train · 13/16CompleteOpen ↗

6.66 TB

PhysInOneP14Train · 14/16CompleteOpen ↗

6.72 TB

PhysInOneP15Train · 15/16CompleteOpen ↗

7.93 TB

PhysInOneP16Train · 16/16CompleteOpen ↗

1.57 TB

Download badges query the official Hugging Face API and update automatically. Counts are repository-level download events, not deduplicated users; accessing multiple shards can produce one event in each shard.

For large-scale rendered-data downloading and filtering, please refer to the Rendered Data section below.

📊 Data Splits

PhysInOne is divided into Train, Val, and Test splits, containing 122,988, 15,411, and 15,411 scenes, respectively. The three splits are generated with completely distinct 3D meshes and backgrounds, ensuring that no scene or visual content is shared across splits. This separation provides a robust benchmark for evaluating generalization and physical reasoning.

🧱 3D Assets

The complete public Unreal Engine 5.5.4 asset release includes the project files and Scene/Trajectory resources for 14,847 validation scenes and 120,660 training scenes. See the dedicated guide for download commands, archive layout, extraction, and project setup.

• Browse the PhysicBenchmark project
• Read the 3D asset download and setup guide

🛠️ Data Processing

The public toolkit covers static and moving camera generation, Level Sequence creation, unattended Unreal Engine rendering, EXR conversion, and multi-view point-cloud generation. The guides describe the Unreal Engine 5.5.4 setup, required project assets, commands, outputs, and safety behavior.

• Browse the data-processing source
• Read the quick-start guide
• Read the complete camera and rendering guide

🏆 Leaderboard Evaluation Data

All user-facing evaluation inputs for the four PhysInOne Leaderboard tasks are publicly available; ground-truth targets remain private. See the dedicated guide for task contents, copyable download commands, manifests, output layout, and resume instructions.

• Browse the public Leaderboard inputs
• Read the Leaderboard download guide

🎞️ Rendered Data

The full rendered dataset is distributed across the 16 repositories listed above. Dependency-free public tools can filter scenes by split, interaction type, and physical phenomenon, then download the selected archives with resume support and per-file logs.

• Read the rendered-data selection and download guide
• Browse the GitHub download tools
• Browse the Hugging Face utility mirror

📦 Dataset Structure

AccelConcaveSpin__bg010__m3Zrtz_trajectory/
├── AccelConcaveSpin__bg010__m3Zrtz_trajectory.json
├── blender_CineCamera_0.json
├── blender_CineCamera_1.json
├── blender_CineCamera_2.json
├── blender_CineCamera_3.json
├── blender_CineCamera_5.json
├── blender_CineCamera_6.json
├── blender_CineCamera_7.json
├── blender_CineCamera_8.json
├── blender_CineCamera_9.json
├── blender_CineCamera_10.json
├── blender_CineCamera_11.json
├── blender_CineCamera_12.json
├── blender_CineCamera_Moving.json
├── cameras.ply
├── caption.txt
├── points3d.ply
├── recorder_stats.json
├── static_camera_list.txt
├── CineCamera_0/
│   ├── depth/  (0000.npz ~ 0089.npz, 90 files)
│   ├── rgb/    (0000.jpg ~ 0089.jpg, 90 files)
│   └── seg/    (0000.npz ~ 0089.npz, 90 files)
├── CineCamera_1/
│   ...
├── CineCamera_2/
│   ...
├── CineCamera_3/
│   ...
├── CineCamera_5/
│   ...
├── CineCamera_6/
│   ...
├── CineCamera_7/
│   ...
├── CineCamera_8/
│   ...
├── CineCamera_9/
│   ...
├── CineCamera_10/
│   ...
├── CineCamera_11/
│   ...
├── CineCamera_12/
│   ...
└── CineCamera_Moving/
    ├── depth/  (0000.npz ~ 0089.npz, 90 files)
    ├── rgb/    (0000.jpg ~ 0089.jpg, 90 files)
    └── seg/    (0000.npz ~ 0089.npz, 90 files)

📝 Annotation Details

PhysInOne provides synchronized visual and physical annotations for each dynamic 3D scene.

•Depth

Depth maps are provided in .npz format and are expressed in meters.

•Segmentation

Segmentation masks encode background, static foreground objects, and dynamic foreground objects.

Expected encoding:

Pixel Value	Meaning
0	Background
1-127	Static foreground objects
128-255	Dynamic foreground objects

•Captions

Each scene includes a caption.txt file containing an English paragraph that describes the visual elements and the physical activity.

•Trajectories

Each sequence includes a JSON file with the same name as the sequence. This file records the per-frame poses of selected dynamic objects, including their rotations and translations in the world coordinate system.

JSON file structure

The JSON file contains two main parts:

• sequence_info: basic information about the sequence, such as the total number of frames and the frame rate (fps).
• actors: pose and metadata for dynamic objects, including object category, asset index path, actor type, and transformation data.

Actor-level pose storage

For each actor, the storage format depends on its actor_type:

• If actor_type is solid, the object is treated as a single rigid body. Its per-frame pose is stored in the actor-level transform_data field, while the components field is empty.
• If actor_type is interactable, the object is represented by multiple components. The actor-level transform_data field is empty, and the per-component poses are stored under components[component_name].transform_data.

Pose entry format in transform_data

Each entry in transform_data describes the pose of the object or component at a specific frame:

• frame: frame index in the sequence
• time_seconds: timestamp of the frame in seconds
• transform: object or component transformation, including:
  • location: 3D position
  • rotation: rotation quaternion
  • scale: 3D scale

•Cameras

Each scene includes multiple camera viewpoints to support 3D perception, reconstruction, and physics-based visual reasoning tasks.

• Taichi-based MPM simulations (elastic solids, plasticine, granular substances, some Newtonian and Non-Newtonian fluids): 15 static cameras + 1 dynamic camera
• Other scenes: 12 static cameras + 1 dynamic camera

Note: Camera IDs are not fixed across cases. The camera index mapping for each case is stored in static_camera_list.txt under that case directory.

Camera JSON structure

Per camera:

• frames: list of per-frame 4×4 camera-to-world (C2W) transformation matrices in Blender coordinate system
• camera_angle_x: horizontal field of view (FOV) in radians, can be used to compute intrinsic focal length
• img_w, img_h: image width and height in pixels
• trajectory_name: name of the camera trajectory
• total_frames, fps: number of frames and frames per second

Dynamic camera behavior

• The dynamic camera moves along a path randomly sampled on a hemisphere surrounding the center of the main objects, providing diverse viewpoints.
• JSON frame indices correspond directly to video frame indices.

•Point Clouds

Each scene should include points.ply.

The initial point cloud was obtained by back-projecting the pixels from each camera view in the first frame based on depth, followed by randomly sampling 100,000 points from the resulting data.

🧩 Data Fields

•Physical Phenomenon and Abbreviations

ID	Physical Phenomenon	Abbreviation	Related Physical Laws
1	Object collide with static, stationary objects	MovingHitsFixed	Laws of Momentum
2	Moving objects collide with non-static stationary objects	MovingHitsStationary	Laws of Momentum
3	Two moving objects collide	MovingHitsMoving	Laws of Momentum
...	...	...	...

Click to expand full abbreviation table

ID	Physical Phenomenon	Abbreviation	Related Physical Laws
4	Objects in equilibrium of wind and gravity	WindGravityBalance	Equilibrium, Aerodynamics, Gravity
5	Wind applied to a stationary object	WindPushStationary	Aerodynamics
6	Wind applied to objects moving in same direction	WindPushSameDir	Aerodynamics
7	Wind applied to objects moving in the opposite direction	WindPushOppDir	Aerodynamics
8	Wind applied to moving objects changes its velocity (applied at an angle)	WindDeflectMotion	Aerodynamics
9	Object thrown up with angle	ObliqueProjectile	Gravity
10	Objects falling straight down	VerticalFall	Gravity
11	Objects rolling down a straight slope	RollDownSlope	Gravity, Friction
12	Objects rolling up a slope	RollUpSlope	Gravity, Friction
13	Magnetic Attraction	MagnetAttract	Magnetism
14	Magnetic Repulsion	MagnetRepel	Magnetism
15	Objects near uniformly rotating pillar	UniformPanelSpin	Laws of Rotation
16	Objects near acceleratingly rotating pillar	AccelPanelSpin	Laws of Rotation
17	Objects inside uniformly rotating bowl	UniformConcaveSpin	Equilibrium, Laws of Rotation, Gravity, Friction
18	Objects inside acceleratingly rotating bowl	AccelConcaveSpin	Equilibrium, Laws of Rotation, Gravity, Friction
19	Objects on uniformly rotating plane	UniformSurfaceSpin	Laws of Rotation, Friction
20	Objects on acceleratingly rotating plane	AccelSurfaceSpin	Laws of Rotation, Friction
21	Objects on coarse surface with friction	FrictionStop	Friction
22	Spring is compressed	SpringCompress	Laws of Elasticity
23	Spring is stretched	SpringStretch	Laws of Elasticity
24	Breakable object shatters	ImpactFracture	Laws of Plasticity
25	Mirror shatters	MirrorFragmentReflect	Laws of Plasticity, Law of Reflection
26	Elastic rope connection	ElasticCouple	Laws of Elasticity
27	Object bounces off spring board	SpringboardRebound	Laws of Elasticity, Laws of Momentum
28	Objects interact with a balanced seesaw	SeesawCenterPivot	Laws of Torque
29	Objects interact with an imbalanced seesaw	SeesawOffsetPivot	Laws of Torque
30	Free balloon floats to ceiling	BalloonFloat	Laws of Buoyancy
31	Tethered balloon pulls string taut	BalloonTether	Laws of Buoyancy, Rope Restraint
32	Multiple balloons lifting	BalloonLift	Laws of Buoyancy, Rope Restraint, Gravity
33	Laser hits flat mirror and reflects	FixedPlanarRedirect	Law of Reflection
34	Laser reflects off multiple mirrors	FixedArrayRedirect	Law of Reflection
35	Laser hits and reflects off concave mirror	FixedConcaveRedirect	Law of Reflection
36	Laser hits and reflects off convex mirror	FixedConvexRedirect	Law of Reflection
37	Mirror sweeps beam	DynMirrorRedirect	Law of Reflection
38	Laser blocked by object	LaserBlock	Light Obstruction
39	Mirror reflection	MirrorReflect	Law of Reflection
40	Cart moving forward with rolling wheels	CartMove	Complex Mechanical Structure Constraints
41	Objects on rotating turntable flies off	RotTurnableInertia	Laws of Rotation, Friction, Laws of Inertia
42	Rotating block pushes another objects	RotBoardInertia	Laws of Inertia, Laws of Rotation, Laws of Momentum
43	One object carrying another	LinCarryInertia	Laws of Inertia, Friction
44	Catapult launches objects	CatapultLaunch	Special Mechanical Structure, Gravity
45	Chain suspends objects	ChainSuspend	Complex Mechanical Structure Constraints, Gravity
46	Objects Swing	SimplePendulum	Laws of Pendulum Motion, Gravity
47	Double Pendulum Moves	DoublePendulum	Laws of Multiple Pendulum Motion, Gravity
48	Crank push objects	CrankPush	Special Mechanical Structure
49	Wall composed of square blocks collapses	BlockWallCollapse	Gravity, Structural Stability
50	Wooden board supported by sticks collapses	StickSupportFail	Gravity, Structural Stability
51	Objects float on the fluid surface	FloatOnLiquid	Laws of Buoyancy, Fluid Dynamics
52	Objects drop into the fluid	DropInLiquid	Laws of Buoyancy, Fluid Dynamics
53	Objects' movement causes fluid motion	MovingObjDriveLiquid	Fluid Dynamics
54	Flowing fluid carries objects along	LiquidCarryMovingObj	Laws of Buoyancy, Fluid Dynamics
55	Fluid flows against stationary objects	LiquidHitFixedObj	Fluid Dynamics
56	Fluid transfers from one container to another	LiquidTransfer	Fluid Dynamics, Conservation of Mass, Surface Tension
57	Fluid passes through several connected containers	LiquidMultiTransfers	Fluid Dynamics, Conservation of Mass, Surface Tension
58	Fluid flows through grid-like structures	LiquidThroughGrid	Fluid Dynamics, Conservation of Mass
59	Fluid moves across mountainous or uneven landscapes	LiquidAcrossUneven	Fluid Dynamics
60	Increasing fluid volume elevates the surface level	LiquidRise	Fluid Dynamics
61	Fluid flows along the contours of an object's surface	LiquidAlongContours	Fluid Dynamics
62	Jet-like fluid projection upward or outward	JetLiquid	Fluid Dynamics
63	Fluid exhibits surface tension	LiquidTension	Fluid Dynamics, Laws of Surface Tension
64	Fluid refracts light when crossing media	LiquidRefraction	Fluid Dynamics, Optics, Law of Refraction (Snell's Law)
65	Sticky fluid drips and accumulates on objects	StickyToObjects	Laws of Cohesion, Viscous Flow
66	Sticky fluid falls from an object's surface	StickyFromObjects	Laws of Cohesion, Laws of Viscous Flow (Navier-Stokes)
67	An elastic object falls and bounces on another surface	ElasticFall	Laws of Elasticity
68	A plasticine object falls and deforms on a surface	PlasticineFall	Laws of Plasticity
69	A Newtonian fluid falls and spreads across a surface	NewtonianFluidFall	Laws of Viscous Flow (Navier-Stokes)
70	A Non-Newtonian fluid falls and flows with variable resistance	NonNewtonianFluidFall	Laws of Viscoplastics Flow
71	A granular substance falls and disperses across a surface	GranularFall	Laws of Friction

Example: MovingHitsStationary_WindDeflectMotion_BalloonFloat__bg140__TZlWl1

• The first three components correspond to specific physical phenomenon abbreviations (see table above).
• bg140 indicates that the scene uses background number 140.
• The last six-character string (TZlWl1) is a unique hash code generated for this scene to guarantee uniqueness.

This naming convention allows users to quickly parse scene metadata and link it to the corresponding physical phenomena, background, and unique identifier.

📜 License

All 3D assets and materials included in PhysInOne have been sourced from publicly available platforms and verified to carry licenses compatible with non-commercial use. These include:

• SketchFab: assets under various licenses, verified that AI-related usage is allowed.
• Fab: assets under CC BY or Unreal Engine Standard License, explicitly permitting AI-related usage.
• BlenderKit: distributed under Royalty-Free (RF) license.
• ShareTextures: textures under CC0 license.

In total, assets comply with licenses including CC BY-NC, CC BY-SA, CC BY-NC-SA, CC0, CC BY, and RF, ensuring all files can be legally used for building a non-commercial dataset. Users must adhere to the original licenses for any redistribution or derivative work.

📚 Citation

If you use PhysInOne in your research, please cite:

@misc{zhou2026physinonevisualphysicslearning,
      title={PhysInOne: Visual Physics Learning and Reasoning in One Suite}, 
      author={Siyuan Zhou and Hejun Wang and Hu Cheng and Jinxi Li and Dongsheng Wang and Junwei Jiang and Yixiao Jin and Jiayue Huang and Shiwei Mao and Shangjia Liu and Yafei Yang and Hongkang Song and Shenxing Wei and Zihui Zhang and Peng Huang and Shijie Liu and Zhengli Hao and Hao Li and Yitian Li and Wenqi Zhou and Zhihan Zhao and Zongqi He and Hongtao Wen and Shouwang Huang and Peng Yun and Bowen Cheng and Pok Kazaf Fu and Wai Kit Lai and Jiahao Chen and Kaiyuan Wang and Zhixuan Sun and Ziqi Li and Haochen Hu and Di Zhang and Chun Ho Yuen and Bing Wang and Zhihua Wang and Chuhang Zou and Bo Yang},
      year={2026},
      eprint={2604.09415},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2604.09415}, 
}

📮 Contact

For questions about the dataset, please contact:

• {siyuan.zhou, hejun.wang, hu123.cheng, jinxi.li}@connect.polyu.hk, bo.yang@polyu.edu.hk