Abstract. Micro-Actions (MAs) are an important form of non-verbal communication in social interactions, with potential applications in human emotional analysis. However, existing methods in Micro-Action Recognition often overlook the inherent subtle changes in MAs, which limits the accuracy of distinguishing MAs with subtle changes. To address this issue, we present a novel Motion-guided Modulation Network (MMN) that implicitly captures and modulates subtle motion cues to enhance spatial-temporal representation learning. Specifically, we introduce a Motion-guided Skeletal Modulation module (MSM) to inject motion cues at the skeletal level, acting as a control signal to guide spatial representation modeling. In parallel, we design a Motion-guided Temporal Modulation module (MTM) to incorporate motion information at the frame level, facilitating the modeling of holistic motion patterns in micro-actions. Finally, we propose a motion consistency learning strategy to aggregate the motion cues from multi-scale features for micro-action classification. Experimental results on the Micro-Action 52 and iMiGUE datasets demonstrate that MMN achieves state-of-the-art performance in skeleton-based micro-action recognition, underscoring the importance of explicitly modeling subtle motion cues.

git clone https://github.com/momiji-bit/MMN
cd MMN

conda create -n MMN python=3.12 -y
conda activate MMN

pip install torch  # cuda version
# pip3 install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu129

pip install scikit-learn tensorboardX timm chardet h5py
pip install -e ./torchlight
pip install -e ./torchpack

Dataset (MA-52 & iMiGUE)

pip install -U huggingface_hub
huggingface-cli login
# export HF_ENDPOINT=https://hf-mirror.com  # (Optional) For users in China, enable the mirror
huggingface-cli download Geo2425/MMN --repo-type dataset --local-dir data

Pretrain Model

• Download Link: here

Training

# Micro-Action 52 Dataset
python main.py --config ./config/train/MA52_J.yaml
python main.py --config ./config/train/MA52_B.yaml
# iMiGUE Dataset
python main.py --config ./config/train/iMiGUE_J.yaml
python main.py --config ./config/train/iMiGUE_B.yaml

Testing

# Micro-Action 52 Dataset
python test.py --config ./config/test/MA52_J.yaml --weights ./path_to_trained_model.pt
python test.py --config ./config/test/MA52_B.yaml --weights ./path_to_trained_model.pt
python test.py --merge ./work_dir/test/MA52_J ./work_dir/test/MA52_B --work-dir ./work_dir/test/MA52_2s
# iMiGUE Dataset
python test.py --config ./config/test/iMiGUE_J.yaml --weights ./path_to_trained_model.pt
python test.py --config ./config/test/iMiGUE_B.yaml --weights ./path_to_trained_model.pt
python test.py --merge ./work_dir/test/iMiGUE_J ./work_dir/test/iMiGUE_B --work-dir ./work_dir/test/iMiGUE_2s

• MA-52: Please submit your test prediction files to the Codabench evaluation server competition page: here

• iMiGUE: Please visit the official MiGA 2025 Challenge website for the latest competition details, timeline, and resources: here.

If you found this work useful, please consider citing:

@article{gu2025motion,
  title={Motion Matters: Motion-guided Modulation Network for Skeleton-based Micro-Action Recognition},
  author={Gu, Jihao and Li, Kun and Wang, Fei and Wei, Yanyan and Wu, Zhiliang and Fan, Hehe and Wang, Meng},
  journal={arXiv preprint arXiv:2507.21977},
  year={2025}
}

@article{guo2024benchmarking,
  title={Benchmarking Micro-action Recognition: Dataset, Methods, and Applications},
  author={Guo, Dan and Li, Kun and Hu, Bin and Zhang, Yan and Wang, Meng},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  year={2024},
  volume={34},
  number={7},
  pages={6238-6252}
}

For any questions, feel free to contact: Dr. Kun Li (kunli.hfut@gmail.com) and Mr. Jihao Gu (jihao.gu.23@ucl.ac.uk).

This code began with SkateFormer. We thank the developers for doing most of the heavy-lifting.