(ICCV 25)MonoFusion: Sparse-View 4D Reconstruction via Monocular Fusion

Project Page | Arxiv | Data

Zihan Wang, Jeff Tan, Tarasha Khurana*, Neehar Peri*, Deva Ramanan

Carnegie Mellon University

* Equal Contribution

Installation

git clone --recurse-submodules https://github.com/ImNotPrepared/MonoFusion
cd MonoFusion/
conda create -n monofusion python=3.10
conda activate monofusion

Update requirements.txt with correct CUDA version for PyTorch and cuUML, i.e., replacing cu122 and cu12 with your CUDA version.

pip install -r requirements.txt
pip install git+https://github.com/nerfstudio-project/gsplat.git

Todo List

Task	Status	Due Date
Drop data and environ build guide	✅ Done	-
Preprocessing scripts	⏳ Todo	in a week
Drop Code	⏳ Todo	between ICLR and ICCV

Usage

We understand that data processing can be complex and time-consuming. Our solution streamlines this workflow - with a single run, you'll receive undistorted, synchronized sparse-view data complete with all necessary priors.

Get Ego-Exo4D Access

To work with Ego-Exo4D data, follow these steps:

1. Obtain License: Get a license for Ego-Exo4D at https://docs.ego-exo4d-data.org/getting-started/

2. Download VRS Files: Download VRS files with RGB stream for a specific take:

   egoexo -o <output_directory> --parts take_vrs --uids <uid1>
   

   Note: VRS files are required to ensure synchronized data streams.

3. Extract Images: Extract undistorted images from the VRS sequences:

   viewer_map --vrs aria01.vrs --task vis --resize 512
   

   The default resize resolution is 512 pixels

Preprocessing

We have one-button-push priors generation script that saves depth maps, human masks, dino feature, and 2D tracks.

./fetch_priors.sh

Start your own EgoRecon (⏳)

Citation

If you find our data, code processing, or project useful, please kindly consider citing our work:

@misc{wang2025monofusionsparseview4dreconstruction,
      title={MonoFusion: Sparse-View 4D Reconstruction via Monocular Fusion}, 
      author={Zihan Wang and Jeff Tan and Tarasha Khurana and Neehar Peri and Deva Ramanan},
      year={2025},
      eprint={2507.23782},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.23782}, 
}