Global-correlated 3D-decoupling Transformer for Clothed Avatar Reconstruction
Authors: Zechuan Zhang, Li Sun, Zongxin Yang, Ling Chen, Yi Yang
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Comprehensive experiments on CAPE and THuman2.0 datasets illustrate that our method outperforms state-of-the-art approaches in both geometry and texture reconstruction, exhibiting high robustness to challenging poses and loose clothing, and producing higher-resolution textures. |
| Researcher Affiliation | Academia | Zechuan Zhang1, Li Sun1, Zongxin Yang1, Ling Chen2, Yi Yang1 1 Re LER, CCAI, Zhejiang University 2 AAII, University of Technology Sydney |
| Pseudocode | No | The paper describes the methodology using prose and diagrams (e.g., Figure 3), but it does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Codes are available at https://github.com/River-Zhang/GTA. |
| Open Datasets | Yes | Our model was trained on the THuman2.0 [15], featuring 526 high-quality human scans, with 505 designated for training and 21 for evaluation. Testing was primarily conducted on the CAPE [16] and THuman2.0, with the former divided into 'CAPE-FP' and 'CAPENFP' subsets to examine model generalization on different pose types. |
| Dataset Splits | No | The paper states that THuman2.0 was used with '505 designated for training and 21 for evaluation', and 'Testing was primarily conducted on the CAPE [16] and THuman2.0'. While it defines training and testing subsets, it does not explicitly mention or provide a separate 'validation' dataset split. |
| Hardware Specification | Yes | The model, implemented in Py Torch Lightning [52], is trained for 10 epochs with a learning rate of 1e-4 and a batch size of 4, over a span of 2 days on a single NVIDIA Ge Force RTX 3090 GPU. |
| Software Dependencies | No | The paper mentions 'implemented in Py Torch Lightning [52]', but it does not provide specific version numbers for PyTorch, PyTorch Lightning, or any other software dependencies crucial for replication. |
| Experiment Setup | Yes | The model, implemented in Py Torch Lightning [52], is trained for 10 epochs with a learning rate of 1e-4 and a batch size of 4, over a span of 2 days on a single NVIDIA Ge Force RTX 3090 GPU. Our 3D-decoupling decoder incorporates both cross-plane and principal-plane decoders, each with a depth of three. Each decoder outputs a feature map F R128 128 64. Following the feature acquisition through our hybrid prior fusion method, two identical Multilayer Perceptrons (MLPs) are employed for separate predictions of occupancy and color, each with layer sizes of [512, 1024, 512, 256, 128, 1]. |