Image-to-3D
PEFT
Safetensors
3D-reconstruction
test-time-adaptation
self-supervised-learning
depth-estimation
3d-vision
Instructions to use PeterDAI/Free-Geometry with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use PeterDAI/Free-Geometry with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
Update README.md
Browse files
README.md
CHANGED
|
@@ -13,4 +13,109 @@ tags:
|
|
| 13 |
- depth-estimation
|
| 14 |
- 3d-vision
|
| 15 |
- peft
|
| 16 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
- depth-estimation
|
| 14 |
- 3d-vision
|
| 15 |
- peft
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
<div align="center">
|
| 19 |
+
|
| 20 |
+
<h1 style="text-align:center;">
|
| 21 |
+
Free Geometry: Refining 3D Reconstruction from Longer Versions of Itself
|
| 22 |
+
</h1>
|
| 23 |
+
|
| 24 |
+
_Test-time self-evolution for feed-forward 3D reconstruction without 3D ground truth_
|
| 25 |
+
|
| 26 |
+
<p align="center">
|
| 27 |
+
<a href="https://arxiv.org/abs/2604.14048" target="_blank">
|
| 28 |
+
<img alt="Paper" src="https://img.shields.io/badge/arXiv-2604.14048-red?logo=arxiv" height="20" />
|
| 29 |
+
</a>
|
| 30 |
+
<a href="https://github.com/hiteacherIamhumble/Free-Geometry" target="_blank">
|
| 31 |
+
<img alt="Code" src="https://img.shields.io/badge/GitHub-Free--Geometry-181717?logo=github" height="20" />
|
| 32 |
+
</a>
|
| 33 |
+
<a href="https://huggingface.co/PeterDAI/Free-Geometry" target="_blank">
|
| 34 |
+
<img alt="Model Weights" src="https://img.shields.io/badge/HuggingFace-Model_Weights-yellow?logo=huggingface" height="20" />
|
| 35 |
+
</a>
|
| 36 |
+
<a href="./LICENSE" target="_blank">
|
| 37 |
+
<img alt="License" src="https://img.shields.io/badge/License-CC_BY_4.0-lightgrey.svg" height="20" />
|
| 38 |
+
</a>
|
| 39 |
+
</p>
|
| 40 |
+
|
| 41 |
+
</div>
|
| 42 |
+
|
| 43 |
+
# Model Description
|
| 44 |
+
|
| 45 |
+
This repository provides a **LoRA / PEFT adapter** for **Free-Geometry** fine-tuning on top of a pretrained base model (VGGT and DepthAnything3).
|
| 46 |
+
|
| 47 |
+
This is **not** a full standalone model checkpoint. The repository only contains the adapter weights and configuration needed to apply the fine-tuned update to the original base model.
|
| 48 |
+
|
| 49 |
+
Use this adapter if you want the Free-Geo fine-tuned behavior while keeping the original base model separate.
|
| 50 |
+
|
| 51 |
+
# What Is In This Repository?
|
| 52 |
+
|
| 53 |
+
This repository is expected to contain:
|
| 54 |
+
|
| 55 |
+
- `adapter_config.json`
|
| 56 |
+
- `adapter_model.safetensors`
|
| 57 |
+
- `README.md`
|
| 58 |
+
|
| 59 |
+
These files define the PEFT adapter and tell PEFT how to attach the LoRA weights to the base model.
|
| 60 |
+
|
| 61 |
+
# Base Model
|
| 62 |
+
|
| 63 |
+
This adapter was trained on top of:
|
| 64 |
+
|
| 65 |
+
- [zoooooooyuan/VGGT-1B](https://huggingface.co/zoooooooyuan/VGGT-1B)
|
| 66 |
+
- [depth-anything/DA3-GIANT-1.1](https://huggingface.co/depth-anything/DA3-GIANT-1.1)
|
| 67 |
+
|
| 68 |
+
|
| 69 |
+
# Training Setup
|
| 70 |
+
|
| 71 |
+
This adapter corresponds to:
|
| 72 |
+
|
| 73 |
+
- Backbone: `DA3` or `VGGT`
|
| 74 |
+
- Method: `Free-Geo`
|
| 75 |
+
- Dataset: `eth3d`, `7scenes`, `hiroom`, or `scannetpp` (all come from [depth-anything/DA3-BENCH](https://huggingface.co/depth-anything/DA3-BENCH)
|
| 76 |
+
|
| 77 |
+
# How To Use With PEFT
|
| 78 |
+
|
| 79 |
+
```python
|
| 80 |
+
from transformers import AutoModel, AutoProcessor
|
| 81 |
+
from peft import PeftModel
|
| 82 |
+
|
| 83 |
+
base_model_id = "YOUR_BASE_MODEL_ID"
|
| 84 |
+
adapter_repo = "YOUR_USERNAME/YOUR_ADAPTER_REPO"
|
| 85 |
+
|
| 86 |
+
processor = AutoProcessor.from_pretrained(base_model_id)
|
| 87 |
+
base_model = AutoModel.from_pretrained(base_model_id)
|
| 88 |
+
model = PeftModel.from_pretrained(base_model, adapter_repo)
|
| 89 |
+
|
| 90 |
+
# Optional: merge adapter into the base model for inference
|
| 91 |
+
# model = model.merge_and_unload()
|
| 92 |
+
```
|
| 93 |
+
|
| 94 |
+
# Repository Variants
|
| 95 |
+
|
| 96 |
+
Free Geometry is designed for test-time adaptation, so you can find different model variants for each datasets.
|
| 97 |
+
|
| 98 |
+
- `DA3-Free-Geo-eth3d`
|
| 99 |
+
- `DA3-Free-Geo-7scenes`
|
| 100 |
+
- `DA3-Free-Geo-hiroom`
|
| 101 |
+
- `DA3-Free-Geo-scannetpp`
|
| 102 |
+
- `VGGT-Free-Geo-eth3d`
|
| 103 |
+
- `VGGT-Free-Geo-7scenes`
|
| 104 |
+
- `VGGT-Free-Geo-hiroom`
|
| 105 |
+
- `VGGT-Free-Geo-scannetpp`
|
| 106 |
+
|
| 107 |
+
# Citation
|
| 108 |
+
|
| 109 |
+
If you use this model, please cite the corresponding project or paper:
|
| 110 |
+
|
| 111 |
+
```bibtex
|
| 112 |
+
@misc{dai2026freegeometryrefining3d,
|
| 113 |
+
title={Free Geometry: Refining 3D Reconstruction from Longer Versions of Itself},
|
| 114 |
+
author={Yuhang Dai and Xingyi Yang},
|
| 115 |
+
year={2026},
|
| 116 |
+
eprint={2604.14048},
|
| 117 |
+
archivePrefix={arXiv},
|
| 118 |
+
primaryClass={cs.CV},
|
| 119 |
+
url={https://arxiv.org/abs/2604.14048},
|
| 120 |
+
}
|
| 121 |
+
```
|