• [March 4, 2026] Accepted to CVPR26.
• [November 22, 2025] Initial release of inference code.

MuM is a Transformer (ViT-L) that extracts highly geometric features from an image. It includes an encoder-decoder structure and for most purposes the decoder is discarded. We show in the paper that MuM beats DINOv3 and CroCo v2 on multi-view tasks such as feedforward 3D reconstruction, matching, and relative pose estimation.

The easiest way to use our trained model is to extract patch features solely through the encoder. We use a similar API as DINOv3, e.g.,

from mum import mum_vitl16
from mum.utils import transform_image

model = mum_vitl16(pretrained=True).cuda().eval()
images = transform_image("assets/wma1.jpg").unsqueeze(0).cuda()
feats = model.forward_features(images)['x_norm_patchtokens'] # B, P, 1024

The full model (including decoder) performs normalized pixel reconstruction in its forward pass. It can also be easily finetuned for any multi-view task. Here is an example use-case:

from mum import mum_vitl16_decoderb
from mum.utils import qualitative_evaluation, transform_image
import torch

IMAGE_LIST = ["assets/wma1.jpg", "assets/wma2.jpg", "assets/wma3.jpg"]
model = mum_vitl16_decoderb(pretrained=True).cuda().eval()
images = torch.stack([transform_image(image_path, size=(256, 256)) for image_path in IMAGE_LIST]).unsqueeze(0).cuda()
# Two example usages:
# (1) Compute reconstruction loss.
loss, pred, mask = model(images, mask_ratio=0.75) # loss should be ~0.23
# (2) Pixel reconstruction.
qualitative_evaluation(model, images, "assets/reconstruction.png")

This should give something like:

[COMING]: We aim to release this soon.

[COMING]: We aim to release all our evaluations in a separate branch.

For this release (solely inference code), you really just need a modern torch and torchvision version. I use 2.6.0+cu118 and 0.21.0+cu118, respectively, together with Python 3.11.9. Using this, I can replicate the reported evaluations (to be released shortly), but any new version should work fine. For visualizations, you also need matplotlib.

To most easily achieve this, you can e.g. run:

conda create -n mum python=3.11.9
conda activate mum

# Any modern version should work fine, this is what I use 
pip install torch==2.6.0+cu118 torchvision==0.21.0+cu118 --extra-index-url https://download.pytorch.org/whl/cu118
# Optional, needed for running code examples
pip install matplotlib

• Provide basic feature extraction inference code
• Release the pre-training code
• Provide data processing scripts and annotation files
• Release all evaluations

Code is MIT licensed. Code for layers is directly descending from DINOv3. DINOv3 has a custom license, see DINOv3.

Our code is largely inspired by VGGT, DINOv3, CroCo, and MAE.

If you find our work useful, please consider citing our paper!

@misc{nordström2025mum,
      title={MuM: Multi-View Masked Image Modeling for 3D Vision}, 
      author={David Nordström and Johan Edstedt and Fredrik Kahl and Georg Bökman},
      year={2025},
      eprint={2511.17309},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.17309}, 
}