Skip to content
Snippets Groups Projects
Commit e39c2d5c authored by Maki94's avatar Maki94
Browse files

docs

parent 96c25ce5
No related merge requests found
......@@ -31,20 +31,29 @@ pip3 install -e git+https://github.com/open-mmlab/mmgeneration@f6551e1d6ca24121d
```
## Static Reconstruction
The project structure follows the origianl 3DGS repository.
### Blender Dataset
To run SplatFields on the Blender dataset, download the origianl [NeRF synthetic dataset](https://drive.google.com/drive/folders/128yBriW1IG_3NJ5Rp7APSTZsJqdJdfc1) and follow the instructions in the `run_blender.sh` script.
Make sure that the `DATASET_ROOT` variable is set to the directory where the Blender dataset is downloaded.
### DTU Dataset
To run our method on the DTU dataset, you could directly download the pre-processed subset released in the [2DGS repo](https://drive.google.com/drive/folders/1SJFgt8qhQomHX55Q4xSvYE2C6-8tFll9).
We provide the bash script `run_dtu.sh` to run SplatFields on the DTU sequences. Update the `DATASET_ROOT` in the script to the path of the downloaded dataset.
## Dynamic Reconstruction
SplatFields straightforwardly extends to dynamic scenes.
### Owlii Dataset
The pre-processed Owlii dataset for the dynamic sparse view reconstruction is available [here](https://drive.google.com/file/d/1OdqwXKmvnpxFI4LC8ckI0eV6r9EZ64ZV/view?usp=sharing) (or download it via `gdown 1OdqwXKmvnpxFI4LC8ckI0eV6r9EZ64ZV && unzip DATA_OWLII`). Then, run the script `run_owlii.sh` to train our model on varying configurations (set the data directory appropriately).
## Citation
If you find our work helpful, please consider citing:
```bibtex
@inproceedings{SplatFields,
title={{SplatFields}: SplatFields: Neural Gaussian Splats for Sparse 3D and 4D Reconstruction},
title={SplatFields: Neural Gaussian Splats for Sparse 3D and 4D Reconstruction},
author={Mihajlovic, Marko and Prokudin, Sergey and Tang, Siyu and Maier, Robert and Bogo, Federica and Tung, Tony and Boyer, Edmond},
booktitle={European Conference on Computer Vision (ECCV)},
year={2024},
......
# Example script to run SplatFields4D on the Owlii dataset
DATASET_ROOT=../ResFields/datasets/public_data
SCENE=dancer_vox11 # exercise_vox11 model basketball
# Download the dataset via `gdown 1OdqwXKmvnpxFI4LC8ckI0eV6r9EZ64ZV && unzip DATA_OWLII`
DATASET_ROOT=./DATA_OWLII
SCENE=dancer # exercise model basketball
N_VIEWS=8 # 10 8 6 4
python train.py -s $DATASET_ROOT/$SCENE --white_background --eval --load_time_step 100 -m ./output_rep/Owlii/${N_VIEWS}views/$SCENE/SplatFields4D --flow_model offset --all_training --train_cam_names cam_train_0 cam_train_1 cam_train_2 cam_train_3 cam_train_4 cam_train_5 cam_train_6 cam_train_7 cam_train_8 cam_train_9 --pts_samples hull --iterations 200000 --encoder_type VarTriPlaneEncoder --num_pts 100000 --num_views 5 --composition_rank 40
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment