Transformer-Based Method for Energy Expenditure Estimation in Sports Science

Strategic Insights -

Researchers at Zhejiang University in Hangzhou, People's Republic of China, have developed a Transformer-based method for estimating energy expenditure (EE) accurately and conveniently in sports science. The method, called Energy Expenditure Estimation Skeleton Transformer (E3SFormer), uses a novel approach that features dual Transformer branches for simultaneous action recognition (AR) and EE regression. The researchers collected 16,526 video clips from 36 participants performing 6 common aerobic exercises and labeled them with continuous calorie readings from COSMED K5. The E3SFormer model yielded a 28.81% mean relative error (MRE) with pure skeleton input, surpassing all comparative models. With multi-modal input, including heart rate and physical attributes, the model achieved a 15.32% MRE, substantially better than other models.

Key Takeaways:

  • Researchers at Zhejiang University proposed a Transformer-based method, E3SFormer, for estimating energy expenditure in sports science.
  • The method uses dual Transformer branches for simultaneous action recognition and EE regression.
  • E3SFormer was trained on 16,526 video clips from 36 participants performing 6 common aerobic exercises.
  • The model yielded a 28.81% mean relative error (MRE) with pure skeleton input, surpassing all comparative models.
  • With multi-modal input, including heart rate and physical attributes, the model achieved a 15.32% MRE, substantially better than other models.
  • The smartwatch showed an 18.10% MRE, indicating room for improvement in contactless measurement for EE.
  • The study is the first attempt to estimate EE using Transformer, promoting contactless and multi-modal physiology analysis for aerobic exercise.
  • The research was published in Frontiers in Physiology, a peer-reviewed journal.

Statistics:

  • 16,526 video clips of 36 participants performing 6 common aerobic exercises.
  • 28.81% mean relative error (MRE) with pure skeleton input.
  • 15.32% MRE with multi-modal input, including heart rate and physical attributes.
  • 18.10% MRE with smartwatch input.
  • 100% of comparative models were surpassed by E3SFormer with pure skeleton input.
  • 100% of comparative models were surpassed by E3SFormer with multi-modal input.

Sources:

  • Zhang, S., et al. "Vision-based multimodal energy expenditure estimation for aerobic exercise in adults." Frontiers in Physiology, 2025;16:1666616.
  • NewsRx. Data on Physiology Reported by Researchers at Zhejiang University (Vision-based multimodal energy expenditure estimation for aerobic exercise in adults). Life Science Weekly. November 4, 2025; p 649.