arXiv:2408.02091v2 Announce Type: replace 
Abstract: 3D human motion prediction aims to generate coherent future motions from observed sequences, yet existing end-to-end regression frameworks often fail to capture complex dynamics and tend to produce temporally inconsistent or static predictions-a limitation rooted in representation shortcutting, where models rely on superficial cues rather than learning meaningful motion structure. We propose a two-stage self-supervised framework that decouples representation learning from prediction. In the pretraining stage, the model performs unified past-future self-reconstruction, reconstructing the past sequence while recovering masked joints in the future sequence under full historical guidance. A velocity-based masking strategy selects highly dynamic joints, forcing the model to focus on informative motion components and internalize the statistical dependencies between past and future states without regression interference. In the fine-tuning stage, the pretrained model predicts the entire future sequence, now treated as fully masked, and is further equipped with a lightweight future-text prediction head for joint optimization of low-level motion prediction and high-level motion understanding. Experiments on Human3.6M, 3DPW, and AMASS show that our method reduces average prediction errors by 8.8% over state-of-the-art methods while achieving competitive future-motion understanding performance compared to LLM-based models. Code is available at: https://github.com/JunyuShi02/MoReFun

تقدم الورقة MoReFun، وهو إطار عمل ذاتي الإشراف من مرحلتين لتوقع حركة الإنسان ثلاثية الأبعاد. يتناول القيود المفروضة على الأطر الحالية التي تعتمد على الانحدار والتي غالبًا ما تنتج تنبؤات غير متسقة من خلال فصل تعلم التمثيل عن التنبؤ. يقوم النموذج أولاً بإعادة بناء التسلسلات الماضية أثناء استعادة المفاصل المستقبلية المخبأة، مع التركيز على المفاصل الديناميكية لتجسيد الاعتماد الإحصائي بين الحالات الماضية والمستقبلية. تهدف هذه الطريقة إلى تعزيز اتساق ودقة توقعات الحركة المستقبلية.

El artículo presenta MoReFun, un marco de aprendizaje auto-supervisado en dos etapas para la predicción del movimiento humano en 3D. Aborda las limitaciones de los marcos de regresión existentes que a menudo producen predicciones inconsistentes al desacoplar el aprendizaje de la representación de la predicción. El modelo reconstruye primero las secuencias pasadas mientras recupera las articulaciones futuras enmascaradas, centrándose en las articulaciones dinámicas para internalizar las dependencias estadísticas entre los estados pasados y futuros. Este enfoque busca mejorar la coherencia y pre…

L'article présente MoReFun, un cadre d'apprentissage auto-supervisé en deux étapes pour la prédiction des mouvements humains en 3D. Il aborde les limites des cadres de régression existants qui produisent souvent des prédictions incohérentes en dissociant l'apprentissage de la représentation de la prédiction. Le modèle reconstruit d'abord les séquences passées tout en récupérant les articulations futures masquées, en se concentrant sur les articulations dynamiques pour intérioriser les dépendances statistiques entre les états passés et futurs. Cette approche vise à améliorer la cohérence et la …

The paper presents MoReFun, a two-stage self-supervised framework for 3D human motion prediction. It addresses the limitations of existing regression frameworks that often produce inconsistent predictions by decoupling representation learning from prediction. The model first reconstructs past sequences while recovering masked future joints, focusing on dynamic joints to internalize statistical dependencies between past and future states. This approach aims to enhance the coherence and accuracy of future motion predictions.

MoReFun: Past-Movement Guided Motion Representation Learning for Future Motion Prediction and Understanding

One More Thing in AI – Your Shortcut to AI Mastery

MoReFun: Past-Movement Guided Motion Representation Learning for Future Motion Prediction and Understanding

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

GPTHumanizer

VibeFrame

sync. labs

AiReelGenerator.com

Ready to build your own newsroom?