Improve model card: Add pipeline tag, links, and usage
#1
by
nielsr
HF Staff
- opened
README.md
CHANGED
|
@@ -1,3 +1,56 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: cc-by-nc-4.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
pipeline_tag: video-to-3d
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# Trace Anything: Representing Any Video in 4D via Trajectory Fields
|
| 7 |
+
|
| 8 |
+
This repository contains the official implementation of the paper [Trace Anything: Representing Any Video in 4D via Trajectory Fields](https://huggingface.co/papers/2510.13802).
|
| 9 |
+
|
| 10 |
+
Trace Anything proposes a novel approach to represent any video as a Trajectory Field, a dense mapping that assigns a continuous 3D trajectory function of time to each pixel in every frame. The model predicts the entire trajectory field in a single feed-forward pass, enabling applications like goal-conditioned manipulation, motion forecasting, and spatio-temporal fusion.
|
| 11 |
+
|
| 12 |
+
Project Page: [https://trace-anything.github.io/](https://trace-anything.github.io/)
|
| 13 |
+
Code: [https://github.com/ByteDance-Seed/TraceAnything](https://github.com/ByteDance-Seed/TraceAnything)
|
| 14 |
+
|
| 15 |
+
## Overview
|
| 16 |
+
<div align="center">
|
| 17 |
+
<img src="https://huggingface.co/depth-anything/trace-anything/resolve/main/assets/teaser.png" width="100%"/>
|
| 18 |
+
</div>
|
| 19 |
+
|
| 20 |
+
## Installation
|
| 21 |
+
For detailed installation instructions, please refer to the [GitHub repository](https://github.com/ByteDance-Seed/TraceAnything#setup).
|
| 22 |
+
|
| 23 |
+
## Sample Usage
|
| 24 |
+
To run inference with the Trace Anything model, first, download the pretrained weights (see GitHub for details). Then, you can use the provided script as follows:
|
| 25 |
+
|
| 26 |
+
```bash
|
| 27 |
+
# Download the model weights to checkpoints/trace_anything.pt
|
| 28 |
+
# Place your input video/image sequence in examples/input/<scene_name>/
|
| 29 |
+
|
| 30 |
+
python scripts/infer.py \
|
| 31 |
+
--input_dir examples/input \
|
| 32 |
+
--output_dir examples/output \
|
| 33 |
+
--ckpt checkpoints/trace_anything.pt
|
| 34 |
+
```
|
| 35 |
+
Results, including 3D control points and confidence maps, will be saved to `<output_dir>/<scene>/output.pt`.
|
| 36 |
+
|
| 37 |
+
## Interactive Visualization
|
| 38 |
+
An interactive 3D viewer is available to explore the generated trajectory fields. Run it using:
|
| 39 |
+
```bash
|
| 40 |
+
python scripts/view.py --output examples/output/<scene>/output.pt
|
| 41 |
+
```
|
| 42 |
+
For more options and remote usage, check the [GitHub repository](https://github.com/ByteDance-Seed/TraceAnything#interactive-visualization-%EF%B8%8F).
|
| 43 |
+
|
| 44 |
+
## Citation
|
| 45 |
+
If you find this work useful, please consider citing the paper:
|
| 46 |
+
```bibtex
|
| 47 |
+
@misc{liu2025traceanythingrepresentingvideo,
|
| 48 |
+
title={Trace Anything: Representing Any Video in 4D via Trajectory Fields},
|
| 49 |
+
author={Xinhang Liu and Yuxi Xiao and Donny Y. Chen and Jiashi Feng and Yu-Wing Tai and Chi-Keung Tang and Bingyi Kang},
|
| 50 |
+
year={2025},
|
| 51 |
+
eprint={2510.13802},
|
| 52 |
+
archivePrefix={arXiv},
|
| 53 |
+
primaryClass={cs.CV},
|
| 54 |
+
url={https://arxiv.org/abs/2510.13802},
|
| 55 |
+
}
|
| 56 |
+
```
|