360Motion Dataset

Dataset of 72,000 immersive 360° videos grouped around 50 entities, 6 Unreal Engine scenes, and 121 typical trajectories. Ideal for 3D video generation or motion synthesis models.

Download dataset

Size

72,000 videos — 9:16 format — 378×672 resolution

Licence

Apache-2.0

Description

‍

360Motion Dataset is a database of 72,000 immersive animated videos in vertical resolution (378×672) designed to train and evaluate motion or video generation models. Each video represents a specific movement or trajectory, divided into 50 entities, 6 virtual scenes using Unreal Engine, and 121 trajectory models.

‍

What is this dataset for?

‍

Train 3D motion generation or animation models
Create AI agents capable of reproducing movements in immersive environments
Optimize video synthesis in vertical formats for mobile or VR

‍

Can it be enriched or improved?

‍

Yes, it is possible to label trajectories or to segment entities according to their nature. The addition of synchronized audio, textual descriptions, or context tagging could reinforce its use in multimodal learning. 3D versions or movement fluidity metrics can also enrich the possible uses.

‍

🔎 In summary

Criterion	Evaluation
🧩 Ease of use	⭐⭐⭐⭐⭐ (Easy – clear structure, videos directly usable)
🧼 Need for cleaning	⭐⭐⭐⭐⭐ (Low – coherent and ready-to-use videos)
🏷️ Annotation richness	⭐⭐⭐✩✩ (Medium – mainly trajectory-based, no semantic description)
📜 Commercial license	✅ Yes (Apache 2.0)
👨‍💻 Beginner friendly	✅ Yes – simple to visualize and integrate into projects
🔁 Fine-tuning ready	✅ Yes – useful for video diffusion, GANs, or imitation learning
🌍 Cultural diversity	⚠️ Neutral – synthetic content without cultural connotation