← ArchivePaper2020
Adult2Child: Motion Style Transfer Using CycleGANs
Yuzhu Dong, Andreas Aristidou, Ariel Shamir, Moshe Mahler, Eakta Jain
CycleGAN on motion words transfers adult mocap into child motion style without temporal alignment of training sequences.
Abstract
Child characters are commonly seen in leading roles in top-selling video games. Previous studies have shown that child motions are perceptually and stylistically different from those of adults. Creating motion for these characters by motion capturing children is uniquely challenging because of confusion, lack of patience and regulations. Retargeting adult motion, which is much easier to record, onto child skeletons, does not capture the stylistic differences. In this paper, we propose that style translation is an effective way to transform adult motion capture data to the style of child motion. Our method is based on CycleGAN, which allows training on a relatively small number of sequences of child and adult motions that do not even need to be temporally aligned. Our adult2child network converts short sequences of motions called motion words from one domain to the other. The network was trained using a motion capture database collected by our team containing 23 locomotion and exercise motions. We conducted a perception study to evaluate the success of style translation algorithms, including our algorithm and recently presented style translation neural networks. Results show that the translated adult motions are recognized as child motions significantly more often than adult motions.
How to read this
- Category
- Method: a motion style transfer technique (CycleGAN)
- Contributions
- Frames adult-to-child motion conversion as unpaired style translation using a CycleGAN trained on short 'motion words'
- Trains on a relatively small, temporally unaligned set of child and adult mocap (a 23-motion locomotion/exercise database collected by the authors)
- Runs a perception study comparing the method against recent style-translation networks
- Context
- Builds on unpaired neural motion style transfer, notably Aberman et al.'s 'Unpaired Motion Style Transfer from Video to Animation', applying the CycleGAN idea to the adult-versus-child stylistic gap.Builds on: Unpaired Motion Style Transfer from Video to Animation
- Correctness
- Rests on the assumption that adult-child differences are a transferable 'style' and that motion-word translation preserves content; validated mainly through a perceptual study on their own modest database, so generalization beyond locomotion/exercise and to other skeletons is unproven.
- Clarity
- Accessible; a first pass conveys the framing and the perception-study outcome, do a second pass for the motion-word representation and CycleGAN losses.
- How to read it
- Focus on the motion-word definition and the perception-study design; a second pass is worth it if you care about how unpaired GAN training avoids temporal alignment.
Builds on
Built upon by
Nothing yet.
Related work
- Automated Extraction and Parameterization of Motions in Large Data Sets 2004 / SIGGRAPH
- Autodesk MotionBuilder 2022 2021 / MotionBuilder
- Motion Warping 1995 / SIGGRAPH
- How to Retarget Motion Capture in MotionBuilder 2019 / MotionBuilder
Keywords
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →