RnG

KO | EN

gitlite — search

#javascript #ai #go #android #rust #linux #docker #java #windows #cli #awesome-list #pytorch #security #open-source

(★ 25)

[CVPR26 Highlight] RnG: A Unified Transformer for Complete 3D Modeling from Partial Observations

RnG 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# RnG: A Unified Transformer for Complete 3D Modeling from Partial Observations

### CVPR 2026 Highlight

<h6>
<a href="https://scholar.google.com/citations?user=OC5oCTgAAAAJ" target="_blank">Mochu Xiang</a>1,2 ‡,
<a href="https://scholar.google.com/citations?user=pfzhh0IAAAAJ" target="_blank">Zhelun Shen</a>2 ✽ †,
<a href="https://scholar.google.com/citations?user=HIeMGxcAAAAJ" target="_blank">Xuesong Li</a>3,4,
<a href="http://npu-cvr.cn/" target="_blank">Jiahui Ren</a>1,
<a href="https://scholar.google.com/citations?user=Qa1DMv8AAAAJ" target="_blank">Jing Zhang</a>3,
 
<a href="https://scholar.google.com/citations?user=kWzyOa8AAAAJ" target="_blank">Chen Zhao</a>2,
<a href="" target="_self"> Shanshan Liu </a>2,
<a href="https://scholar.google.com/citations?user=pnuQ5UsAAAAJ" target="_blank">Haocheng Feng</a>2,
<a href="https://jingdongwang2017.github.io" target="_blank">Jingdong Wang</a>2,
<a href="https://scholar.google.com/citations?user=fddAbqsAAAAJ" target="_blank">Yuchao Dai</a>1 †
</h6>
 1 Northwestern Polytechnical University, China
 2 Baidu Inc., China 
 3 Australian National University, Australia 
 4 CSIRO, Australia

<div align="center">
 <a href="https://npucvr.github.io/RnG/">Project Page</a> |
 <a href="https://arxiv.org/abs/2603.01194">Paper</a> |
 <a href="https://huggingface.co/MochuXiang/RnG/tree/main">Model</a> |
 <a href="https://huggingface.co/datasets/MochuXiang/FluffyElephant/tree/main">Data_44k</a> |
 <a href="https://huggingface.co/datasets/benzlxs/objaverse_rendering_set/tree/main">Data_80k</a>
</div>

</div>

> **The Network Architecture of RnG.** (a) Source view images are first tokenized using the DINO vision transformer; the Plücker ray map representing the target view point goes through a linear layer. After adding camera tokens for each view, all tokens will then alternately attend to global- and frame-level attention blocks. Finally, camera tokens from input views are used to estimate camera poses, while a point head and an RGB head process ray tokens from the target view, providing geometry and appearance estimations. (b) In inference, the model can cache K/V token from source views, synthesizing novel view geometry and geometry at a higher speed.

## 1. Run the demo

### 1.1 Environment

``` 
conda create -n rng python=3.11
conda activate rng
pip install -r requirements.txt
```

### 1.2 Checkpoints

Download the checkpoint `RnG.pt` from [Huggingface](https://huggingface.co/MochuXiang/RnG/tree/main) and put it under `experiments/checkpoints/RnGUP_Med`.

### 1.3 Evaluation & Demo data

Download the mluti-view renderings of the GSO dataset `gso_render_25v.tar` from [Huggingface](https://huggingface.co/datasets/MochuXiang/FluffyElephant/tree/main), extract it and modify the path in `configs/RnG_obj_medium_bf16_40k.yaml`:
``` 
 val_dataset_cfgs:
 root_dir: /workspace/ssd1/gso_render_rv # <- modify line 82
 suffix: render_mvs_25/model/ # <- 
```

### 1.4 Launch the demo
```
bash scripts/run_viser_demo.sh
```

## 2. Train the model

We are organizing our training code. Meanwhile, you can checkout the training data [LVIS_subset_44k_25v](https://huggingface.co/datasets/MochuXiang/FluffyElephant/tree/main) and [LGM_subset_80k_40v](https://huggingface.co/datasets/benzlxs/objaverse_rendering_set/tree/main). They contain posed multi-view RGBD renderings of the Objaverse dataset.

RnG

RnG 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

RnG

RnG 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README