KwaiVGI/360Motion-Dataset

360°-Motion Dataset

Project page | Paper | Code

Acknowledgments

We thank Jinwen Cao, Yisong Guo, Haowen Ji, Jichao Wang, and Yi Wang from Kuaishou Technology for their help in constructing our 360°-Motion Dataset.

News

• [2024-12] We release the V1 dataset (72,000 videos consists of 50 entities, 6 UE scenes, and 121 trajectory templates).

Data structure

 ├── 360Motion-Dataset                      Video Number        Cam-Obj Distance (m)
   ├── 480_720/384_672
       ├── Desert (desert)                    18,000               [3.06, 13.39]
           ├── location_data.json
       ├── HDRI                                                      
           ├── loc1 (snowy street)             3,600               [3.43, 13.02]
           ├── loc2 (park)                     3,600               [4.16, 12.22]
           ├── loc3 (indoor open space)        3,600               [3.62, 12.79]
           ├── loc11 (gymnastics room)         3,600               [4.06, 12.32]
           ├── loc13 (autumn forest)           3,600               [4.49  11.91]
           ├── location_data.json
       ├── RefPic
       ├── CharacterInfo.json
       ├── Hemi12_transforms.json

(1) Released Dataset Information

Argument	Description	Argument	Description
Video Resolution	(1) 480×720 (2) 384×672	Frames/Duration/FPS	99/3.3s/30
UE Scenes	6 (1 desert+5 HDRIs)	Video Samples	(1) 36,000 (2) 36,000
Camera Intrinsics (fx,fy)	(1) 1060.606 (2) 989.899	Sensor Width/Height (mm)	(1) 23.76/15.84 (2) 23.76/13.365
Hemi12_transforms.json	12 surrounding cameras	CharacterInfo.json	entity prompts
RefPic	50 animals	1/2/3 Trajectory Templates	36/60/35 (121 in total)
{D/N}_{locX}	{Day/Night}_{LocationX}	{C}_ {XX}_{35mm}	{Close-Up Shot}_{Cam. Index(1-12)} _{Focal Length}

Note that the resolution of 384×672 refers to our internal video diffusion resolution. In fact, we render the video at a resolution of 378×672 (aspect ratio 9:16), with a 3-pixel black border added to both the top and bottom.

(2) Difference with the Dataset to Train on Our Internal Video Diffusion Model

The release of the full dataset regarding more entities and UE scenes is 1) still under our internal license check, 2) awaiting the paper decision.

Argument	Released Dataset	Our Internal Dataset
Video Resolution	(1) 480×720 (2) 384×672	384×672
Entities	50 (all animals)	70 (20 humans+50 animals)
Video Samples	(1) 36,000 (2) 36,000	54,000
Scenes	6	9 (+city, forest, asian town)
Trajectory Templates	121	96

(3) Load Dataset Sample

1. Change root path to dataset. We provide a script to load our dataset (video & entity & pose sequence) as follows. It will generate the sampled video for visualization in the same folder path.

   python load_dataset.py
   

2. Visualize the 6DoF pose sequence via Open3D as follows.

   python vis_trajecotry.py
   

   After running the visualization script, you will get an interactive window like this.

   

Citation

@article{fu20243dtrajmaster,
        author  = {Fu, Xiao and Liu, Xian and Wang, Xintao and Peng, Sida and Xia, Menghan and Shi, Xiaoyu and Yuan, Ziyang and Wan, Pengfei and Zhang, Di and Lin, Dahua},
        title   = {3DTrajMaster: Mastering 3D Trajectory for Multi-Entity Motion in Video Generation},
        journal = {arXiv preprint arXiv:2412.07759},
        year    = {2024}
    }

Contact

Xiao Fu: [email protected]