FourierHandFlow: Neural 4D Hand Representation Using Fourier Query Flow

Authors: Jihyun Lee, Junbong Jang, Donghwan Kim, Minhyuk Sung, Tae-Kyun Kim

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In the experiments, our method achieves state-of-the-art results on video-based 4D reconstruction while being computationally more efficient than the existing 3D/4D implicit shape representations. In the experiments, we validate the effectiveness of FOURIERHANDFLOW on video-based 4D hand reconstruction using Inter Hand2.6M [26] dataset, where we achieve state-of-the-art results in comparison to the existing (1) image-based 3D hand shape reconstruction methods and (2) 4D implicit shape reconstruction methods modified to take RGB hand sequences as inputs.
Researcher Affiliation Academia 1KAIST 2Imperial College London {jyun.lee, junbongjang, kdoh2522, mhsung, kimtaekyun}@kaist.ac.kr
Pseudocode No The paper includes a block diagram (Figure 2) but no formal pseudocode or algorithm block.
Open Source Code Yes Our code is available at https://github.com/jyunlee/Fourier Hand Flow.
Open Datasets Yes We use the two-hand (TH) and single-hand (SH) subsets of 30 FPS version Inter Hand2.6M [26] dataset, which contains diverse hand motions captured in RGB sequences with dense shape annotations.
Dataset Splits Yes For each subset, we use samples annotated as valid hand type and follow the train/val/test splits of the original Inter Hand2.6M dataset. The resulting TH subset contains 477K training and 4K validation sequences, and SH subset contains 656K training and 5K validation sequences.
Hardware Specification No The paper mentions computational efficiency and inference time but does not provide specific hardware details (e.g., GPU/CPU models) used for running the experiments.
Software Dependencies No The paper mentions using 'PyTorch [31] library' but does not specify its version or the versions of any other software dependencies.
Experiment Setup No The paper states 'For more details on network training and architecture, please refer to the supplementary section.', indicating that specific experimental setup details are not provided in the main text.