Add model card for INSPATIO-WORLD
#1
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,3 +1,59 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
--
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
pipeline_tag: image-to-video
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# INSPATIO-WORLD: A Real-Time 4D World Simulator via Spatiotemporal Autoregressive Modeling
|
| 7 |
+
|
| 8 |
+
INSPATIO-WORLD is a real-time framework capable of recovering and generating high-fidelity, dynamic interactive scenes from a single reference video. It utilizes a Spatiotemporal Autoregressive (STAR) architecture to ensure global consistency and precise interaction control.
|
| 9 |
+
|
| 10 |
+
[[Paper](https://huggingface.co/papers/2604.07209)] [[Project Page](https://inspatio.github.io/inspatio-world/)] [[GitHub](https://github.com/inspatio/inspatio-world)]
|
| 11 |
+
|
| 12 |
+
## Installation
|
| 13 |
+
|
| 14 |
+
To set up the environment, follow these steps:
|
| 15 |
+
|
| 16 |
+
```bash
|
| 17 |
+
conda env create -f environment.yml
|
| 18 |
+
conda activate inspatio_world
|
| 19 |
+
|
| 20 |
+
# Install flash-attn
|
| 21 |
+
pip install https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
|
| 22 |
+
```
|
| 23 |
+
|
| 24 |
+
## Inference
|
| 25 |
+
|
| 26 |
+
The full pipeline runs in three steps: video captioning, depth estimation, and spatiotemporal autoregressive inference. You can run the full pipeline using the provided script:
|
| 27 |
+
|
| 28 |
+
### Quick Start
|
| 29 |
+
```bash
|
| 30 |
+
# 1. Place your .mp4 video(s) in a folder
|
| 31 |
+
mkdir -p my_videos
|
| 32 |
+
cp your_video.mp4 my_videos/
|
| 33 |
+
|
| 34 |
+
# 2. Run the full pipeline (using 14B model)
|
| 35 |
+
bash run_test_pipeline.sh \
|
| 36 |
+
--input_dir ./my_videos \
|
| 37 |
+
--traj_txt_path ./traj/x_y_circle_cycle.txt
|
| 38 |
+
```
|
| 39 |
+
|
| 40 |
+
### Using the 1.3B Model
|
| 41 |
+
For a lighter and faster version, use the 1.3B model:
|
| 42 |
+
```bash
|
| 43 |
+
bash run_test_pipeline.sh \
|
| 44 |
+
--input_dir ./test/example \
|
| 45 |
+
--traj_txt_path ./traj/x_y_circle_cycle.txt \
|
| 46 |
+
--config_path ./configs/inference_1.3b.yaml \
|
| 47 |
+
--checkpoint_path ./checkpoints/InSpatio-World-1.3B/InSpatio-World-1.3B.safetensors
|
| 48 |
+
```
|
| 49 |
+
|
| 50 |
+
## Citation
|
| 51 |
+
|
| 52 |
+
```bibtex
|
| 53 |
+
@misc{inspatio-world,
|
| 54 |
+
title={InSpatio-World},
|
| 55 |
+
author={InSpatio Team},
|
| 56 |
+
howpublished={\url{https://github.com/inspatio/inspatio-world}},
|
| 57 |
+
year={2026}
|
| 58 |
+
}
|
| 59 |
+
```
|