3D Human Pose Estimation via Explicit Compositional Depth Maps
Authors: Haiping Wu, Bin Xiao12378-12385
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show our method achieves superior performance on large-scale 3D pose datasets Human3.6M and MPI-INF-3DHP, and sets the new state-of-the-art.We evaluate our method and perform ablation study on the large-scale Human3.6M dataset (Ionescu et al. 2014), and show generalization ability on the in-the-wild MPI-INF-3DHP benchmark (Mehta et al. 2017a) and 2D MPII (Andriluka et al. 2014) test set. |
| Researcher Affiliation | Collaboration | Haiping Wu,1 Bin Xiao2 1Mc Gill University, 2Byde Dance AI Lab haiping.wu2@mail.mcgill.ca, xiaobin.ailab@bytedance.com |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described, nor does it contain a statement of release or a link to a repository. |
| Open Datasets | Yes | We evaluate our method and perform ablation study on the large-scale Human3.6M dataset (Ionescu et al. 2014), and show generalization ability on the in-the-wild MPI-INF-3DHP benchmark (Mehta et al. 2017a) and 2D MPII (Andriluka et al. 2014) test set. |
| Dataset Splits | Yes | Following previous works, we split the dataset into training set of five subjects (S1, S2, S3, S4, S5) and test set of two subjects (S9, S11). Every 5 frames of the training set are used for training, while we test on every 64 frames of test set following (Sun et al. 2018). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, processor types, or memory amounts) used for running its experiments, only mentioning the use of specific networks like ResNet50 and HRNet-W32. |
| Software Dependencies | No | The paper mentions software components like 'Simple Baselines', 'HRNet', and 'Adam optimizer' but does not provide specific version numbers for these or any other ancillary software dependencies (e.g., programming language versions, library versions). |
| Experiment Setup | Yes | We train all the methods for 20 epochs using Adam optimizer, with of initial learning rate of 0.001, and decreases 10 times at the 15th, 17th epochs. Rotation, synthetic occlusion (S ar andi et al. 2018) and Photo metric distortion are used as data augmentation. The input size of images is 256 x 256. |