Add pipeline tag: image-to-3d
#1
by
nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,66 +1,67 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
--
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
[.
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
pipeline_tag: image-to-3d
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# AnySplat: Feed-forward 3D Gaussian Splatting from Unconstrained Views
|
| 7 |
+
|
| 8 |
+
[](https://city-super.github.io/anysplat/)
|
| 9 |
+
[](https://arxiv.org/pdf/2505.23716)
|
| 10 |
+
[](https://github.com/OpenRobotLab/AnySplat)
|
| 11 |
+
[](https://huggingface.co/lhjiang/anysplat)
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
## Quick Start
|
| 15 |
+
|
| 16 |
+
See the Github repository: https://github.com/OpenRobotLab/AnySplat regarding installation instructions.
|
| 17 |
+
|
| 18 |
+
The model can then be used as follows:
|
| 19 |
+
|
| 20 |
+
```python
|
| 21 |
+
from pathlib import Path
|
| 22 |
+
import torch
|
| 23 |
+
import os
|
| 24 |
+
import sys
|
| 25 |
+
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
| 26 |
+
|
| 27 |
+
from src.misc.image_io import save_interpolated_video
|
| 28 |
+
from src.model.model.anysplat import AnySplat
|
| 29 |
+
from src.utils.image import process_image
|
| 30 |
+
|
| 31 |
+
# Load the model from Hugging Face
|
| 32 |
+
model = AnySplat.from_pretrained("anysplat_ckpt_v1")
|
| 33 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 34 |
+
model = model.to(device)
|
| 35 |
+
model.eval()
|
| 36 |
+
for param in model.parameters():
|
| 37 |
+
param.requires_grad = False
|
| 38 |
+
|
| 39 |
+
# Load and preprocess example images (replace with your own image paths)
|
| 40 |
+
image_names = ["path/to/imageA.png", "path/to/imageB.png", "path/to/imageC.png"]
|
| 41 |
+
images = [process_image(image_name) for image_name in image_names]
|
| 42 |
+
images = torch.stack(images, dim=0).unsqueeze(0).to(device) # [1, K, 3, 448, 448]
|
| 43 |
+
b, v, _, h, w = images.shape
|
| 44 |
+
|
| 45 |
+
# Run Inference
|
| 46 |
+
gaussians, pred_context_pose = model.inference((images+1)*0.5)
|
| 47 |
+
|
| 48 |
+
pred_all_extrinsic = pred_context_pose['extrinsic']
|
| 49 |
+
pred_all_intrinsic = pred_context_pose['intrinsic']
|
| 50 |
+
save_interpolated_video(pred_all_extrinsic, pred_all_intrinsic, b, h, w, gaussians, image_folder, model.decoder)
|
| 51 |
+
|
| 52 |
+
```
|
| 53 |
+
|
| 54 |
+
## Citation
|
| 55 |
+
|
| 56 |
+
```
|
| 57 |
+
@article{jiang2025anysplat,
|
| 58 |
+
title={AnySplat: Feed-forward 3D Gaussian Splatting from Unconstrained Views},
|
| 59 |
+
author={Jiang, Lihan and Mao, Yucheng and Xu, Linning and Lu, Tao and Ren, Kerui and Jin, Yichen and Xu, Xudong and Yu, Mulin and Pang, Jiangmiao and Zhao, Feng and others},
|
| 60 |
+
journal={arXiv preprint arXiv:2505.23716},
|
| 61 |
+
year={2025}
|
| 62 |
+
}
|
| 63 |
+
```
|
| 64 |
+
|
| 65 |
+
## License
|
| 66 |
+
|
| 67 |
The code and models are licensed under the [MIT License](LICENSE).
|