Neural SDF Flow for 3D Reconstruction of Dynamic Scenes

Authors: Wei Mao, Richard Hartley, Mathieu Salzmann, miaomiao Liu

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on real-world multi-view video datasets show that our reconstructions are better than those of the state-of-the-art methods.
Researcher Affiliation Collaboration Australian National University wei.mao@anu.edu.au Richard Hartley Australian National University & Google richard.hartley@anu.edu.au Mathieu Salzmann CVLab, EPFL & SDSC, Switzerland mathieu.salzmann@epfl.ch Miaomiao Liu Australian National University miaomiao.liu@anu.edu.au
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/ wei-mao-2019/SDFFlow.git.
Open Datasets Yes We evaluate our method quantitativly on the CMU Panoptic dataset (Joo et al., 2017) and qualitatively on the Tensor4D dataset (Shao et al., 2023).
Dataset Splits No The paper does not explicitly provide training/validation/test dataset splits with percentages or sample counts for reproducibility of model training, but states 'we use all 10 camera views for training and only evaluate the meshes'.
Hardware Specification Yes We train our model for 2000 epochs, which takes around 7 days on 2 NVIDIA 4090 GPUs for ten 1920 1080 videos of 24 frames.
Software Dependencies No We implement our method using Pytorch (Paszke et al., 2017) and use the Adam optimizer (Kingma & Ba, 2014) to train our model.
Experiment Setup Yes The batch size is set to 1024. We use the second-order Runge-Kutta method to solve the integration in Equation 5. We train our model for 2000 epochs... The balancing weight λ is set to 0.1.