Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
[SIGGRAPH Asia 2026] AnyRecon: Arbitrary-View 3D Reconstruction with Video Diffusion Model
| Date | Stars |
|---|---|
| 2026-07-31 | 388 |
| 2026-08-06 | 389 |
| 2026-08-14 | 391 |
| 2026-08-18 | 394 |
| 2026-08-19 | 395 |
| 2026-08-22 | 396 |
| 2026-08-26 | 397 |
| 2026-08-27 | 398 |
| 2026-08-31 | 399 |
| 2026-09-03 | 400 |
| 2026-09-05 | 401 |
| 2026-09-11 | 402 |
| 2026-09-15 | 403 |
| 2026-09-20 | 403 |
Today
— stars today
This week
+1 stars this week
This month
+8 stars this month
Momentum
0.0
growth rate 0.25%/day
<p align="center" >
<img src="docs/logo.png" width="60%" >
</p>
<h2 align="center">AnyRecon: Arbitrary-View 3D Reconstruction<br>with Video Diffusion Model</h2>
<div align="center">
<a href="https://yutian10.github.io">Yutian Chen</a>,
<a href="https://guoshi28.github.io">Shi Guo</a>,
<a href="https://rbjin.github.io/">Renbiao Jin</a>,
<a href="https://scholar.google.com/citations?user=9b5dE40AAAAJ&hl=en">Tianshuo Yang</a>,
<a href="https://caixin98.github.io/">Xin Cai</a>,
<a href="https://luo0207.github.io/yawenluo/">Yawen Luo</a>,
<a href="">Mingxin Yang</a>, <br>
<a href="https://mulinyu.github.io/">Mulin Yu</a>,
<a href="https://eveneveno.github.io/lnxu/">Linning Xu</a>,
<a href="https://tianfan.info/">Tianfan Xue</a>
</div>
<br>
<p align="center">
<a href="https://arxiv.org/pdf/2604.19747"><img src="https://img.shields.io/static/v1?label=Arxiv&message=AnyRecon&color=red&logo=arxiv"></a>
<a href="https://yutian10.github.io/AnyRecon/" target="_blank">
<img src="https://img.shields.io/badge/Project%20Page-Website-3273dc?logo=googlechrome&logoColor=white">
</a>
<a href='https://huggingface.co/Yutian10/AnyRecon'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-yellow'></a>
<a href='https://youtu.be/sfgFZKCdofs'><img src='https://img.shields.io/badge/YouTube-Video-FF0000?logo=youtube&logoColor=white'></a>
</p>
<h3 align="center"> SIGGRAPH Asia 2026 </h3>
<br>
<p align="center">
<b>Your star means a lot for us to develop this project! ✨</b>
</p>
<p align="center" width="100%">
<img src="docs/gif.gif" width="90%" >
</p>
## TODO List
- [ ] Upload sparse attention weight.
## 🛠️ Environment Setup
### 1. Clone Repository and Setup Environment
The point-cloud rendering pipeline depends on [π³](https://github.com/yyfz/Pi3/), which is included as a git submodule. Make sure to clone **recursively** so that `Pi3/` is fetched at the same time:
```bash
git clone --recursive https://github.com/OpenImagingLab/AnyRecon.git
# If you already cloned without --recursive, run:
# git submodule update --init --recursive
cd AnyRecon
conda create -n anyrecon python=3.10 -y
conda activate anyrecon
pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt
pip install -r Pi3/requirements.txt
```
### 2. Download Models
AnyRecon relies on specific pre-trained weights. Please download them and place them in the `./checkpoints` folder.
- Base Video Diffusion Model (Wan2.1 I2V 14B 720P) [[download](https://huggingface.co/Wan-AI/Wan2.1-I2V-14B-720P/tree/main)]
- AnyRecon LoRA weights [[download](https://huggingface.co/Yutian10/AnyRecon/tree/main)]
- π³ checkpoint (for point-cloud rendering) [[download](https://huggingface.co/yyfz233/Pi3/resolve/main/model.safetensors)] → place at `Pi3/model.safetensors`
## 🚀 Quick Start
For inference, processing an 869x512 video at 40 frames requires approximately 45GB of VRAM, but you can lower the resolution if your VRAM is insufficient. To reproduce the provided example, run:
```bash
bash test.sh
```
Or directly:
```bash
python run_AnyRecon.py \
--root_dir example/valley \
--output_dir example/valley \
--lora_path full_attention.ckpt
```
## 🌟 Run on Your Own Data
`run_AnyRecon.py` expects point-cloud rendered **condition videos** as input. To prepare them from a raw video, we provide a helper script built on top of [π³](https://github.com/yyfz/Pi3/):
```bash
bash run_pi3.sh
```
**Input video format.** Your input video must be organized so that:
- the **first `--num_cond_frames` frames** are the **capture views** — these provide the 3D point cloud,
- the **remaining frames** are the **test views** — they are *only* used to estimate the camera poses at which the point cloud is rendered, and **do not contribute any points** to the reconstruction.
**Custom test-view trajectory (no test frames needed).**Excerpt of 5,255 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:eb2d60adece62733, desc:3d reconstruction