NeuroFluid: Fluid Dynamics Grounding with Particle-Driven Neural Radiance Fields

Authors: Shanyan Guan, Huayu Deng, Yunbo Wang, Xiaokang Yang

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4. Experiments
Researcher Affiliation Academia 1Mo E Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University, Shanghai 200240, China.
Pseudocode No The paper describes the methods in prose and with equations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code and video demonstrations are available at https: //syguan96.github.io/Neuro Fluid.
Open Datasets No We train Neuro Fluid on visual observations of fluids generated by DFSPH (Bender & Koschier, 2015) and Blender (Community, 2018). We generate four benchmarks named Honey Cone , Water Cube , Water Sphere , and Water Bunny .
Dataset Splits No The paper specifies using 'the visual observations during the first 50 time steps as the training data' but does not provide explicit training, validation, and test dataset splits with percentages or counts.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or cloud instance types used for running the experiments.
Software Dependencies No The paper mentions using the Adam optimizer (Kingma & Ba, 2015) and building upon Ne RF (Mildenhall et al., 2020), but it does not specify version numbers for any software libraries or dependencies (e.g., PyTorch, TensorFlow, CUDA).
Experiment Setup Yes In the warm-up phase of Rϕ, it is trained for 100k steps with an initial learning rate of 5e 4, which is then exponentially decayed with γ = 0.1. In the joint training phase, the entire model is trained for 500k steps. The initial learning rate of Tθ and Rϕ are set to 1e 6 and 5e 4 respectively. The learning rate of Tθ is decayed by 0.5 after 10k, 30k, 50k, 100k, and 300k steps. That of Rϕ is decayed by 0.5 after 10k, 75k, and 150k steps. The resolution of the observed image is 400 400.