Add dataset card and metadata for InTraGen
#2
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
task_categories:
|
| 3 |
+
- text-to-video
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# InTraGen: Trajectory-controlled Video Generation for Object Interactions
|
| 7 |
+
|
| 8 |
+
This repository contains the datasets introduced in the paper [InTraGen: Trajectory-controlled Video Generation for Object Interactions](https://huggingface.co/papers/2411.16804).
|
| 9 |
+
|
| 10 |
+
InTraGen is a pipeline for improved trajectory-based generation of object interaction scenarios. The researchers introduced four new datasets to evaluate performance in these contexts:
|
| 11 |
+
|
| 12 |
+
- **MoVi-Extended**
|
| 13 |
+
- **Domino**
|
| 14 |
+
- **Football**
|
| 15 |
+
- **Pool**
|
| 16 |
+
|
| 17 |
+
The official code and further instructions can be found in the [GitHub repository](https://github.com/insait-institute/intragen).
|
| 18 |
+
|
| 19 |
+
## Citation
|
| 20 |
+
|
| 21 |
+
If you find these datasets useful for your research, please consider citing the following paper:
|
| 22 |
+
|
| 23 |
+
```bibtex
|
| 24 |
+
@article{liu2024intragen,
|
| 25 |
+
title={InTraGen: Trajectory-controlled Video Generation for Object Interactions},
|
| 26 |
+
author={Liu, Zuhao and Yanev, Aleksandar and Mahmood, Ahmad and Nikolov, Ivan and Motamed, Saman and Zheng, Wei-Shi and Wang, Xi and Van Gool, Luc and Paudel, Danda Pani},
|
| 27 |
+
journal={arXiv preprint arXiv:2411.16804},
|
| 28 |
+
year={2024}
|
| 29 |
+
}
|
| 30 |
+
```
|