Toward Approaches to Scalability in 3D Human Pose Estimation

Authors: Jun-Hui Kim, Seong-Whan Lee

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our results demonstrate that these approaches increase the diversity and volume of pose data while consistently achieving performance gains, even amid the complexities introduced by increased pose diversity.Our experiments were conducted on the Human3.6M (H36M) [20], MPI-INF-3DHP (3DHP) [21], and 3DPW [22] benchmarks.
Researcher Affiliation Academia Jun-Hee Kim Seong-Whan Lee Dept. of Artificial Intelligence, Korea University, Seoul 02841, Republic of Korea {jh__kim, sw.lee}@korea.ac.kr
Pseudocode Yes Algorithm 1 Biomechanical Pose Generator (BPG) and Algorithm 2 Binary Depth Coordinates (BDC) are provided in Appendix D.
Open Source Code Yes The code can be found in the supplementary material.
Open Datasets Yes Our experiments were conducted on the Human3.6M (H36M) [20], MPI-INF-3DHP (3DHP) [21], and 3DPW [22] benchmarks.
Dataset Splits No The paper mentions using training data and evaluation, but does not explicitly specify a distinct validation dataset split with percentages or counts.
Hardware Specification Yes For the BPG, training is conducted on a single Nvidia GTX 3090 Ti GPU with a batch size of 1024 for 100 epochs, taking approximately two days to complete the experiments.
Software Dependencies No The paper mentions software components like VPose [24], Adam optimizer, and HRnet [37], but does not provide specific version numbers for these or other libraries.
Experiment Setup Yes We train the models on the H36M dataset for 50 epochs with a batch size of 1024. All components are optimized using the Adam optimizer with an initial learning rate of 0.001, which linearly decays over time.