File size: 1,838 Bytes
a959425 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 |
---
license: cc-by-nc-sa-4.0
---
# 4DGT Model Card
## Model Details
4DGT (4D Gaussian Transformer) is a neural network model that learns dynamic 3D Gaussian representations from monocular videos. It uses a transformer-based architecture to predict 4D Gaussians from a dynamic scenes observed from an egocentric video.
- **Paper:** [4DGT: Learning a 4D Gaussian Transformer Using Real-World Monocular Videos](https://arxiv.org/abs/2506.08015)
- **Project Page:** [https://4dgt.github.io/](https://4dgt.github.io/)
- **Github:** [GitHub repository](https://github.com/facebookresearch/4dgt)
Please refer to the project page and github for more details of the model.
## Citation
```bibtex
@inproceedings{xu20254dgt,
title = {4DGT: Learning a 4D Gaussian Transformer Using Real-World Monocular Videos},
author = {Xu, Zhen and Li, Zhengqin and Dong, Zhao and Zhou, Xiaowei and Newcombe, Richard and Lv, Zhaoyang},
journal = {arXiv preprint arXiv:2506.08015},
year = {2025}
}
```
## Model Files
### Checkpoint: `4dgt_full.pth`
- **Size:** ~14.5 GB
- **Format:** PyTorch state dict
- **Contents:**
- The full model trained as described in the paper.
- Encoder weights (DINOv2 backbone)
- Level of Details Transformer
- 4D Gaussian Decoder
### Checkpoint: `4dgt_1st_stage.pth`
- **Size:** ~4.85 GB
- **Format:** PyTorch state dict
- **Contents:**
- The first stage model trained only using Egoexo4D dataset as described in the paper.
- Encoder weights (DINOv2 backbone)
- Vanilla Transformer, no level of details.
- 4D Gaussian Decoder
## Quick Start
Please refer to [4DGT GitHub repository](https://github.com/facebookresearch/4dgt) for the full set up.
## Contact
For questions and issues, please open an issue on the [GitHub repository](https://github.com/facebookresearch/4dgt). |