Self-Supervised 3D Human Mesh Recovery from a Single Image with Uncertainty-Aware Learning
Authors: Guoli Yan, Zichun Zhong, Jing Hua
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on benchmark datasets show that our approach outperforms other state-of-the-art methods at similar supervision levels. We conduct extensive experiments on benchmark datasets and achieve state-of-the-art results against previous methods at similar supervision levels. In our experiments, we use Human3.6M (Ionescu et al. 2013), 3DPW (Von Marcard et al. 2018) and UP-3D (Lassner et al. 2017) for training. Ablation Study In Table 1, we compare our proposed model with several variants on 3DPW dataset to investigate the contribution of each component. |
| Researcher Affiliation | Academia | Department of Computer Science, Wayne State University, Detroit, MI, USA {guoliyan, zichunzhong, jinghua}@wayne.edu |
| Pseudocode | No | The paper describes its methods in narrative text and provides a system overview diagram (Figure 1), but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | In our experiments, we use Human3.6M (Ionescu et al. 2013), 3DPW (Von Marcard et al. 2018) and UP-3D (Lassner et al. 2017) for training. |
| Dataset Splits | Yes | For 3DPW (Von Marcard et al. 2018) and Human3.6M (Ionescu et al. 2013)... The dataset has 60 video sequences. Both training and testing sets contain 24 videos, and the rest 12 video are used for validation. Following the Protocol 2 (Kanazawa et al. 2018), we train our model on 5 subjects (S1, S5, S6, S7, S8) and test on the front-view samples of the rest 2 subjects (S9, S11). |
| Hardware Specification | Yes | Our experiments run on a single NVIDIA Ge Force RTX 3090 GPU. |
| Software Dependencies | No | The paper mentions using the Adam optimizer and ResNet-18, but it does not specify software dependencies with version numbers (e.g., Python, PyTorch/TensorFlow, CUDA versions). |
| Experiment Setup | Yes | The MLP network for uncertainty modeling contains two fully connected layers with 128 and 64 neurons, respectively. The output layer contains 12 neurons, which is equal to the number of joints excluding joints on the head. We set the joint loss weight αj = 1 and the depth loss weight αd = 0.04. We use the Adam optimizer (Kingma and Ba 2014) with an initial learning rate of 10 5, and batch size of 64. After training for 10 epochs, we regularize β to remain close to the mean shape. |