Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,87 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: cc-by-sa-4.0
|
| 3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-sa-4.0
|
| 3 |
+
---
|
| 4 |
+
# MultiScene360 Dataset
|
| 5 |
+
**A Real-World Multi-Camera Video Dataset for Generative Vision AI**
|
| 6 |
+
|
| 7 |
+
## π Overview
|
| 8 |
+
The MultiScene360 Dataset is designed to advance generative vision AI by providing synchronized multi-camera footage from real-world environments.
|
| 9 |
+
|
| 10 |
+
π‘ **Key Applications**:
|
| 11 |
+
β Video generation & view synthesis
|
| 12 |
+
β 3D reconstruction & neural rendering
|
| 13 |
+
β Digital human animation systems
|
| 14 |
+
β Virtual/augmented reality development
|
| 15 |
+
|
| 16 |
+
## π Dataset Specifications (Public Version)
|
| 17 |
+
| Category | Specification |
|
| 18 |
+
|---------------------|----------------------------------------|
|
| 19 |
+
| Scenes | 10 base + 3 extended scenes |
|
| 20 |
+
| Scene Duration | 10-20 seconds each |
|
| 21 |
+
| Camera Views | 4 synchronized angles per scene |
|
| 22 |
+
| Total Video Clips | ~144 |
|
| 23 |
+
| Data Volume | 20-30GB (1080p@30fps) |
|
| 24 |
+
|
| 25 |
+
*Commercial version available with 200+ scenes and 6-8 camera angles*
|
| 26 |
+
|
| 27 |
+
## π Complete Scene Specification
|
| 28 |
+
| ID | Environment | Location | Primary Action | Special Features |
|
| 29 |
+
|------|-------------|-----------------|----------------------------|---------------------------|
|
| 30 |
+
| S001 | Indoor | Living Room | Walk β Sit | Occlusion handling |
|
| 31 |
+
| S002 | Indoor | Kitchen | Pour water + Open cabinet | Fine hand motions |
|
| 32 |
+
| S003 | Indoor | Corridor | Walk β Turn | Depth perception |
|
| 33 |
+
| S004 | Indoor | Desk | Type β Head turn | Upper body motions |
|
| 34 |
+
| S005 | Outdoor | Park | Walk β Sit (bench) | Natural lighting |
|
| 35 |
+
| S006 | Outdoor | Street | Walk β Stop β Phone check | Gait variation |
|
| 36 |
+
| S007 | Outdoor | Staircase | Ascend stairs | Vertical movement |
|
| 37 |
+
| S008 | Indoor | Corridor | Two people passing | Multi-person occlusion |
|
| 38 |
+
| S009 | Indoor | Mirror | Dressing + mirror view | Reflection surfaces |
|
| 39 |
+
| S010 | Indoor | Empty room | Dance movements | Full-body dynamics |
|
| 40 |
+
| S011 | Indoor | Window | Phone call + clothes adjust| Silhouette + semi-reflections |
|
| 41 |
+
| S012 | Outdoor | Shopping street | Walking + window browsing | Transparent surfaces + crowd |
|
| 42 |
+
| S013 | Indoor | Night corridor | Walking + light switching | Low-light adaptation |
|
| 43 |
+
|
| 44 |
+
## π₯ Camera Configuration
|
| 45 |
+
**Physical Setup**:
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
|
| 50 |
+
**Technical Details**:
|
| 51 |
+
- **Cameras**: DJI Osmo Action 5 Pro (4 identical units)
|
| 52 |
+
- **Mounting**: Tripod-stabilized at ~1.5m height
|
| 53 |
+
- **Distance**: 2-3m from subject center
|
| 54 |
+
- **FOV Overlap**: 20-30% between adjacent cameras
|
| 55 |
+
|
| 56 |
+
## π Suggested Research Directions
|
| 57 |
+
1. Cross-view consistency learning
|
| 58 |
+
2. Novel view synthesis from sparse inputs
|
| 59 |
+
3. Dynamic scene reconstruction
|
| 60 |
+
4. Human motion transfer between viewpoints
|
| 61 |
+
|
| 62 |
+
## π Access Information
|
| 63 |
+
π― **[Sample Download Here](https://madacode.file.core.windows.net/root/360/detaset_sample_part.zip?sv=2023-01-03&st=2025-05-06T08%3A56%3A56Z&se=2028-01-07T08%3A56%3A00Z&sr=f&sp=r&sig=5R2FrdBqw35HIF0r2TaUxAsr0mz5h7oKDUHFFpkD8ik%3D)**
|
| 64 |
+
|
| 65 |
+
β¨ **Free Full Dataset Download**: [https://maadaa.ai/multiscene360-Dataset](https://maadaa.ai/multiscene360-Dataset)
|
| 66 |
+
|
| 67 |
+
πΌ **Commercial Inquiries**: [contact@maadaa.ai](mailto:contact@maadaa.ai)
|
| 68 |
+
|
| 69 |
+
**Usage Rights:**
|
| 70 |
+
β Free for academic/commercial use
|
| 71 |
+
β License: Attribution-NonCommercial-ShareAlike 4.0 International
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
## About maadaa.ai
|
| 75 |
+
Founded in 2015, maadaa.ai is a pioneering AI data service provider specializing in multimodal data solutions for generative AI development. We deliver end-to-end data services covering text, voice, image, and video datatypes β the core fuel for training and refining generative models.
|
| 76 |
+
|
| 77 |
+
**Our Generative AI Data Solution includes:**
|
| 78 |
+
κ· High-quality dataset collection & annotation tailored for LLMs and diffusion models
|
| 79 |
+
κ· Scenario-based human feedback (RLHF/RLAIF) to enhance model alignment
|
| 80 |
+
κ· One-stop data management through our MaidX platform for streamlined model training
|
| 81 |
+
|
| 82 |
+
**Why Choose Us**:
|
| 83 |
+
β Reduce real-world data collection costs by 70%+
|
| 84 |
+
β Generate perfectly labeled training data at scale
|
| 85 |
+
β API-first integration for synthetic pipelines
|
| 86 |
+
|
| 87 |
+
*Empowering the next generation of interactive media and spatial computing**
|