datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Physics-aware-videos中文 README
🧱 Physics-aware Video Dataset
This dataset is a high-quality real-world video dataset focused on physical phenomena, designed for learning and evaluating physical laws from videos. It primarily covers the following classic physical processes:
🧱 Rigid-body motion / 🌊 Fluid dynamics
💫 Collision and rebound
💥 Explosion and burst phenomena
🌫️ Smoke, dust, and particle scattering
🌍 Gravity and inertia (falling, rolling, acceleration, etc.)
The dataset is constructed… See the full description on the dataset page: https://huggingface.co/datasets/BAAI-DataCube/Physics-aware-videos.physics-scenarios-packed
physics-scenarios-packed
Packed (tar.gz) version of a 2D rigid body physics dataset for training language models on next-frame prediction. 1,000,020 scenes × 200 frames simulated with Pymunk / Chipmunk2D.
This repo is bandwidth-friendly: each scenario type ships as a single .tar.gz. For the unpacked JSONL files see physics-scenarios-raw.
Scale
Train: 900,000 scenes (24 seen scenario types × 37,500 each)
Val: 100,020 scenes (30 scenario types × 3,334 each — includes 6… See the full description on the dataset page: https://huggingface.co/datasets/AlexWortega/physics-scenarios-packed.handbook-chemistry-physics-tablesphysics-scenarios-raw
physics-scenarios-raw
Raw (un-tarred) JSONL version of the 2D rigid body physics dataset. Each scene is a separate file under <split>/<scenario_type>/scene_<id>.jsonl. Streaming-friendly for HF datasets and curriculum sampling.
For the bandwidth-efficient packaged version, see physics-scenarios-packed.
Scale (this snapshot)
Train: 80,000 scenes
Val: 10,000 scenes
Test: 10,000 scenes
Frames per scene: 200
Format: one .jsonl per scene (1 header + 200 frame lines)
This… See the full description on the dataset page: https://huggingface.co/datasets/AlexWortega/physics-scenarios-raw.