Invariant Teacher and Equivariant Student for Unsupervised 3D Human Pose Estimation
Authors: Chenxin Xu, Siheng Chen, Maosen Li, Ya Zhang3013-3021
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on Human3.6M and MPI-INF3DHP. Our method reduces the 3D joint prediction error by 11.4% compared to state-of-the-art unsupervised methods and also outperforms many weakly-supervised methods that use side information on Human3.6M. Code will be available at https://github.com/sjtuxcx/ITES. |
| Researcher Affiliation | Academia | Cooperative Medianet Innovation Center, Shanghai Jiao Tong University {xcxwakaka, sihengc, maosen_li, ya_zhang}@sjtu.edu.cn |
| Pseudocode | No | The paper does not contain pseudocode or a clearly labeled algorithm block. |
| Open Source Code | No | Code will be available at https://github.com/sjtuxcx/ITES. |
| Open Datasets | Yes | Human3.6M. Human3.6M (Ionescu et al. 2013) is a large-scale dataset used widely in 3D pose estimation. MPI-INF-3DHP. We also use another large scale 3D human pose dataset, MPI-INF-3DHP (Mehta et al. 2017), which includes poses in both indoor and outdoor scenes. |
| Dataset Splits | No | The paper states 'we consider the poses of the subjects S1, S5, S6, S7, and S8 for training, and use S9 and S11 for testing' but does not explicitly mention a validation split or how it's handled. |
| Hardware Specification | No | The paper does not specify the hardware used for the experiments (e.g., CPU, GPU models, or cloud computing resources). |
| Software Dependencies | No | The paper mentions using an 'SGD (Bottou 2010) optimizer' but does not provide specific software names with version numbers for dependencies like Python, PyTorch, TensorFlow, etc. |
| Experiment Setup | Yes | We set the distance between the camera and the root joint (pelvis) t as 5 unit. In the teacher network, we set the size of the pose dictionary K to 12. In the student network, we use 8 graph convolution blocks and a residual connection is built across consecutive two blocks. We train the entire framework with the SGD (Bottou 2010) optimizer. In the first training stage, we train the teacher network with the learning rate 0.001 for 40 epoches. In the second training stage, we train the student network for 30 epoches with the learning rate 0.001. The weight parameter is set as λREP = 5, λRIC = 1, λKD = 5, λREC = 1. |