Continuous Piecewise-Affine Based Motion Model for Image Animation
Authors: Hexiang Wang, Fengqi Liu, Qianyu Zhou, Ran Yi, Xin Tan, Lizhuang Ma
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on four datasets demonstrate the effectiveness of our method against state-of-the-art competitors quantitatively and qualitatively. |
| Researcher Affiliation | Collaboration | Hexiang Wang1 , Fengqi Liu1 , Qianyu Zhou1*, Ran Yi1 , Xin Tan2, Lizhuang Ma1,2 1 Department of Computer Science and Engineering, Shanghai Jiao Tong University, Shanghai, China 2 East China Normal University, Shanghai, China {whxsjtu123, liufengqi, zhouqianyu, ranyi}@sjtu.edu.cn, xtan@cs.ecnu.edu.cn, ma-lz@cs.sjtu.edu.cn. This work was also supported by Shenzhen Smart More Corporation, Ltd, which provided GPUs and computing resources for us. |
| Pseudocode | Yes | The whole calculating process is summarized in Algorithm 1 in supplementary. |
| Open Source Code | No | Code will be publicly available at: https://github.com/Devil PG/AAAI2024-CPABMM. |
| Open Datasets | Yes | Tai Chi HD (Siarohin et al. 2019b) consists of videos showcasing full-body Tai Chi performances downloaded from You Tube and cropped to a resolution of 256 256 based on the bounding boxes around the performers. TED-talks (Siarohin et al. 2021) contains videos of TED talks downloaded from You Tube, which were downscaled to 384 384 resolution based on the upper human bodies. The video length ranges from 64 to 1024 frames. Vox Celeb (Nagrani et al. 2017) comprises of videos featuring various celebrities speaking, which were downloaded from You Tube and cropped to a resolution of 256 256 based on the bounding boxes surrounding their faces. Video length ranges from 64 to 1024 frames. MGif (Siarohin et al. 2019a) is a collection of .gif files featuring pixel animations of animals in motion, which was obtained through Google searches. |
| Dataset Splits | No | We use the same data pre-processing protocol and train-test split strategy as in (Siarohin et al. 2021). The paper refers to an external source for the split strategy and does not provide specific details on validation splits or percentages within its own text. |
| Hardware Specification | Yes | We used one Ge Force RTX 3090 GPU to train our model for 100 epochs in all datasets with an initial learning rate of 0.0001. |
| Software Dependencies | No | We implement the framework with Py Torch and use Adam optimizer to update our model. The paper mentions PyTorch but does not provide its version number or any other software dependencies with version numbers. |
| Experiment Setup | Yes | We used one Ge Force RTX 3090 GPU to train our model for 100 epochs in all datasets with an initial learning rate of 0.0001. ... We set the training hyper-parameters as: λr = 10, λe = 10, λk = 1, λs = 0.1. |