Multi-Scale Control Signal-Aware Transformer for Motion Synthesis without Phase

Authors: Lintao Wang, Kun Hu, Lei Bai, Yu Ding, Wanli Ouyang, Zhiyong Wang

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Both qualitative and quantitative experimental results on an existing biped locomotion dataset, which involves diverse types of motion transitions, demonstrate the effectiveness of our method.
Researcher Affiliation Collaboration 1School of Computer Science, The University of Sydney, Australia 2Shanghai AI Laboratory, China 3Netease Fuxi AI Lab, China
Pseudocode No The paper describes its methods using text and mathematical equations but does not include any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement about releasing source code or a link to a code repository for the described methodology.
Open Datasets Yes We evaluate our proposed method on a public dataset (Holden, Komura, and Saito 2017) for a fair comparison with the state-of-the-art methods.
Dataset Splits No The paper mentions using 'around 4 million samples for training' but does not specify the splits for training, validation, or testing sets, nor does it provide percentages or counts for these splits.
Hardware Specification Yes The model was trained with 20 epochs, which took 50 hours on an NVIDIA GTX 1080Ti GPU.
Software Dependencies Yes The model was implemented by Py Torch 1.7.1 (Paszke et al. 2019) and trained with an Adam optimisier (Kingma and Ba 2014).
Experiment Setup Yes In total, K = 5 past frames with indices k1 = 1, k2 = 10, k3 = 20, k4 = 30 and k5 = 40 were selected as input to predict the motion of the i-th frame. ... Each of them consisted of three transformer-encoder layers using six self-attention heads of a dimension 186 and the the feed-forward layers were of a dimension 1024. A dropout rate of 0.1 was applied to the encoders. ... The motion prediction network was modelled as a three-layer MLP with a hidden dimension 512 and a dropout rate 0.3. ... λ for ℓ1 regularization was set to 0.01. The learning rate was set to 10 4 and the batch size was 32. The model was trained with 20 epochs