ReliaAvatar: A Robust Real-Time Avatar Animator with Integrated Motion Prediction

Authors: Bo Qian, Zhenhuan Wei, Jiashuo Li, Xing Wei

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4 Experiments 4.1 Implementation Details 4.2 Evaluation Metrics 4.3 Evaluation Protocols 4.4 Comparison to the State-of-the-art 4.5 Ablation Studies
Researcher Affiliation Academia Bo Qian, Zhenhuan Wei, Jiashuo Li, Xing Wei School of Software Engineering, Xi an Jiaotong University {qb990531, zh-wei, xjtuljs}@stu.xjtu.edu.cn,weixing@mail.xjtu.edu.cn
Pseudocode No The paper describes the model architecture and training process but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes The code is available at https://github.com/MIV-XJTU/Relia Avatar.
Open Datasets Yes We compare Relia Avatar with existing methods in all scenarios on the AMASS benchmark dataset [Mahmood et al., 2019]. Following the experimental setup outlined in Avatar Poser [Jiang et al., 2022], we partition the CMU [graphics lab., 2000], BMLrub [Troje, 2002], and HDM05 [M uller et al., 2007] subsets into 90% training data and 10% testing data.
Dataset Splits No Section 4.3 "Evaluation Protocols" states: "...we partition the CMU [graphics lab., 2000], BMLrub [Troje, 2002], and HDM05 [M uller et al., 2007] subsets into 90% training data and 10% testing data." This explicitly mentions training and testing data splits but does not specify a validation split.
Hardware Specification Yes The model is trained on two Ge Force RTX 3090 GPUs for a total of 90000 iterations, with a batch size of 32 on each GPU. All tests are conducted on a single Ge Force RTX 3090 GPU.
Software Dependencies No The paper describes the use of GRU and Transformer modules, but it does not specify any software versions for frameworks (e.g., PyTorch, TensorFlow) or programming languages used (e.g., Python version).
Experiment Setup Yes For all GRUs in the model, the number of layers is set to 1 and the dimension is set to 256. The joint-relation Transformer has 4 layers with a dimension of 512. The initial learning rate is set to 5 10 4, and is halved every 15000 iterations. The length of the input sequence L is set to 32. {λori, λrot, λSMP L pos , λdec pos, λvec} = {0.02, 1, 1, 1, 0.5}. The model is trained on two Ge Force RTX 3090 GPUs for a total of 90000 iterations, with a batch size of 32 on each GPU.