MoRe Pretrained Weights

This repository is an unofficial mirror of the pretrained weights released for MoRe: Motion-aware Feed-forward 4D Reconstruction Transformer. The checkpoints are intended to be used with the official MoRe codebase for model inference.

Source and provenance

The files in this repository were copied, without modification, from the Google Drive folder published by the MoRe authors.

This mirror is provided only as an alternative download location. It is not affiliated with, endorsed by, or maintained by the original MoRe authors. Please refer to the official repository, the project page, and the paper for authoritative documentation, licensing, usage terms, and updates.

Files

File	Description	Size	SHA-256
more_full.pt	Full-attention MoRe checkpoint	5,677,547,294 bytes	8ac67b4870da62052a7529d9c71db200d943f9b061f71459c721665ca97454d0
more_stream.pt	Streaming/grouped-causal-attention MoRe checkpoint	5,677,555,805 bytes	7db3af5ec6b2a977e2fe24aa7789a713d26c607214fad1bfb1e70054bc358dac

Usage

Clone and install the official MoRe repository:

git clone https://github.com/HellexF/MoRe
cd MoRe

conda create -n more python=3.10 -y
conda activate more
conda install pytorch=2.9.0 torchvision=0.24.0 cudatoolkit=11.8 -c pytorch
conda install cudatoolkit-dev=11.8 -c conda-forge
pip install -r requirements.txt

Download the mirrored checkpoints into the directory expected by MoRe:

hf download haikuoxin/more --local-dir pretrained

Run inference with the full-attention checkpoint:

python inference.py \
    --config_path training/config/omniworld_full.yaml \
    --ckpt_path pretrained/more_full.pt \
    --image_path ./data/example_video \
    --output_dir ./results/full_res \
    --conf_thres 50.0 \
    --predict_motion

For streaming inference with more_stream.pt, follow the official repository's MagiAttention installation and streaming configuration instructions.

About MoRe

MoRe is a feed-forward 4D reconstruction transformer for recovering dynamic 3D scenes from monocular videos. The official project describes two core components: motion–structure disentanglement and grouped causal attention.

Citation

If you use these weights, please cite the original MoRe work:

@inproceedings{fang2026moremotionawarefeedforward4d,
  title     = {MoRe: Motion-aware Feed-forward 4D Reconstruction Transformer},
  author    = {Juntong Fang and Zequn Chen and Weiqi Zhang and Donglin Di and
               Xuancheng Zhang and Chengmin Yang and Yu-Shen Liu},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year      = {2026}
}