File size: 2,808 Bytes
edc8c98
d767379
edc8c98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0ab439a
edc8c98
 
 
 
 
 
 
 
d767379
 
 
 
 
 
 
edc8c98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
---
license: cc-by-nc-4.0
library_name: pytorch
pipeline_tag: image-to-3d
tags:
  - gaussian-splatting
  - novel-view-synthesis
  - feed-forward-3dgs
  - 3d-reconstruction
  - ecosplat
---

# EcoSplat: Efficiency-controllable Feed-forward 3D Gaussian Splatting from Multi-view Images

**CVPR 2026 (Highlight)**

[Paper](https://arxiv.org/abs/2512.18692) · [Project Page](https://kaist-viclab.github.io/ecosplat-site/) · [Code](https://github.com/KAIST-VICLab/EcoSplat)

EcoSplat is a feed-forward 3D Gaussian Splatting model whose compute/quality trade-off is **controllable at inference time** via a single primitive budget, the protect rate κ (lower κ → fewer rendered Gaussians). This repository hosts the efficiency-controllable (IGF) checkpoints.

## Checkpoints

| File | Base | Train data |
| --- | --- | --- |
| `ecosplat-spfsplat-re10k.ckpt` | SPFSplat | RealEstate10K |
| `ecosplat-zpressor-re10k.ckpt` | ZPressor / MVSplat | RealEstate10K |

## License & provenance

Released under **CC BY-NC 4.0** (non-commercial; see `LICENSE`). © 2026 Minh-Quan Viet Bui.

- `ecosplat-spfsplat-re10k.ckpt` is fine-tuned from **MASt3R** (Naver), which is non-commercial (CC BY-NC-SA 4.0) — that restriction applies to this checkpoint.
- `ecosplat-zpressor-re10k.ckpt` is fine-tuned from **MVSplat + ZPressor** (MIT).

## Usage

Download a checkpoint:

```python
from huggingface_hub import hf_hub_download
ckpt = hf_hub_download("quan5609/EcoSplat", "ecosplat-spfsplat-re10k.ckpt")
```

or

```bash
wget https://huggingface.co/quan5609/EcoSplat/resolve/main/ecosplat-spfsplat-re10k.ckpt
```

Then follow the inference / evaluation instructions for the matching base, on the corresponding code branch:

- **SPFSplat base** — [`eco_spfsplat`](https://github.com/KAIST-VICLab/EcoSplat/tree/eco_spfsplat)
- **ZPressor / MVSplat base** — [`eco_zpressor`](https://github.com/KAIST-VICLab/EcoSplat/tree/eco_zpressor)

Set the primitive budget at test time with `+model.encoder.igf.inference_rho=<κ>` (e.g. sweep `0.7 → 0.1 → 0.02`).

## Citation

```bibtex
@article{park2025ecosplat,
  title={EcoSplat: Efficiency-controllable Feed-forward 3D Gaussian Splatting from Multi-view Images},
  author={Park, Jongmin and Bui, Minh-Quan Viet and Bello, Juan Luis Gonzalez and Moon, Jaeho and Oh, Jihyong and Kim, Munchurl},
  journal={arXiv preprint arXiv:2512.18692},
  year={2025}
}
```

## Acknowledgements

Built on [SPFSplat](https://github.com/ranrhuang/SPFSplat), [ZPressor](https://github.com/ziplab/ZPressor), [NoPoSplat](https://github.com/cvg/NoPoSplat), [pixelSplat](https://github.com/dcharatan/pixelsplat), [MVSplat](https://github.com/donydchen/mvsplat), [DepthSplat](https://github.com/cvg/depthsplat), [DUSt3R](https://github.com/naver/dust3r), and [CroCo](https://github.com/naver/croco).