d4rt 0.0.5

Implementation of D4RT, Efficiently Reconstructing Dynamic Scenes

d4rt (wip)

Implementation of D4RT, Efficiently Reconstructing Dynamic Scenes, Deepmind

install

$ pip install d4rt

usage

from torch import randn, randint
from d4rt.d4rt import D4RT

model = D4RT(
    dim = 512,
    video_image_size = 128,
    video_patch_size = 32,
    video_max_time_len = 10,
    enc_depth = 6,
    dec_depth = 6
)

videos = randn(2, 10, 3, 128, 128)

video_lens = randint(2, 10, (2,)) # handle variable lengthed video, can be None for max length always

# inputs

coors = randint(0, 128, (2, 5, 2))
time_src = randint(0, 10, (2, 5))
time_tgt = randint(0, 10, (2, 5))
time_camera = randint(0, 10, (2, 5))

query_lens = randint(1, 5, (2,)) # handle varaible lengthed queries

# output

points = randn(2, 5, 3)

loss = model(
    videos,
    video_lens = video_lens,
    coors = coors,
    time_src = time_src,
    time_tgt = time_tgt,
    time_camera = time_camera,
    query_lens = query_lens,
    points = points,
)

loss.backward()

# without giving the output, it returns the prediction

pred = model(
    videos,
    coors = coors,
    time_src = time_src,
    time_tgt = time_tgt,
    time_camera = time_camera
)

assert pred.shape == (2, 5, 3)

citations

@article{zhang2025d4rt,
    title   = {Efficiently Reconstructing Dynamic Scenes One D4RT at a Time},
    author  = {Zhang, Chuhan and Le Moing, Guillaume and Koppula, Skanda and Rocco, Ignacio and Momeni, Liliane and Xie, Junyu and Sun, Shuyang and Sukthankar, Rahul and Barral, Jo{\"e}lle K. and Hadsell, Raia and Ghahramani, Zoubin and Zisserman, Andrew and Zhang, Junlin and Sajjadi, Mehdi S. M.},
    journal = {arXiv preprint},
    year    = {2025}
}

@inproceedings{liu2026geometryaware,
    title   = {Geometry-aware 4D Video Generation for Robot Manipulation},
    author  = {Zeyi Liu and Shuang Li and Eric Cousineau and Siyuan Feng and Benjamin Burchfiel and Shuran Song},
    booktitle = {The Fourteenth International Conference on Learning Representations},
    year    = {2026},
    url     = {https://openreview.net/forum?id=18gC6pZVVc}
}

@misc{joseph2026interpretingphysicsvideoworld,
    title   = {Interpreting Physics in Video World Models},
    author  = {Sonia Joseph and Quentin Garrido and Randall Balestriero and Matthew Kowal and Thomas Fel and Shahab Bakhtiari and Blake Richards and Mike Rabbat},
    year    = {2026},
    eprint  = {2602.07050},
    archivePrefix = {arXiv},
    primaryClass = {cs.CV},
    url={https://arxiv.org/abs/2602.07050},
}