mlsanityData to 14 Oct 2026

Motion Style Transfer: Modular Low-Rank Adaptation for Deep Motion Forecasting

arXiv 2211.03165 ↗First posted 6 Nov 2022cs.CV cs.ROcited by 12

Not on the map. Why?

Abstract (arXiv)

Deep motion forecasting models have achieved great success when trained on a massive amount of data. Yet, they often perform poorly when training data is limited. To address this challenge, we propose a transfer learning approach for efficiently adapting pre-trained forecasting models to new domains, such as unseen agent types and scene contexts. Unlike the conventional fine-tuning approach that updates the whole encoder, our main idea is to reduce the amount of tunable parameters that can precisely account for the target domain-specific motion style. To this end, we introduce two components that exploit our prior knowledge of motion style shifts: (i) a low-rank motion style adapter that projects and adjusts the style features at a low-dimensional bottleneck; and (ii) a modular adapter strategy that disentangles the features of scene context and motion history to facilitate a fine-grained choice of adaptation layers. Through extensive experimentation, we show that our proposed adapter design, coined MoSA, outperforms prior methods on several forecasting benchmarks.

Key facts. Cited by 12 papers in the map (data through 14 Oct 2026). Builds on 31 of its 45 references that are in the map. First posted on arXiv 6 Nov 2022.

Builds on

31 of 45 references.

References · 31

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  2. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  3. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,897
  4. 2019Parameter-Efficient Transfer Learning for NLPcited by 4,236
  5. 2013Distributed Representations of Words and Phrases and their Compositionalitycited by 3,474
  6. 2022Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context Learningcited by 896
  7. 2016COCO-Stuff: Thing and Stuff Classes in Contextcited by 864
  8. 2018Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networkscited by 690
  9. 2018Deep Visual Domain Adaptation: A Surveycited by 635
  10. 2021Compacter: Efficient Low-Rank Hypercomplex Adapter Layerscited by 460
  11. 2020Social-STGCNN: A Social Spatio-Temporal Graph Convolutional Neural Network for Human Trajectory Predictioncited by 325
  12. 2016Revisiting Batch Normalization For Practical Domain Adaptationcited by 288
19 more
  1. 2018Residual Reinforcement Learning for Robot Controlcited by 274
  2. 2020One Thousand and One Hours: Self-driving Motion Prediction Datasetcited by 240
  3. 2017Social Attention: Modeling Attention in Human Crowdscited by 186
  4. 2018Efficient parametrization of multi-domain deep neural networkscited by 168
  5. 2017Modulating early visual processing by languagecited by 167
  6. 2020From Goals, Waypoints & Paths To Long Term Human Trajectory Forecastingcited by 167
  7. 2019The inD Dataset: A Drone Dataset of Naturalistic Road User Trajectories at German Intersectionscited by 137
  8. 2020Human Trajectory Forecasting in Crowds: A Deep Learning Perspectivecited by 129
  9. 2020Deep Visual Domain Adaptationcited by 91
  10. 2020Social NCE: Contrastive Learning of Socially-aware Motion Representationscited by 58
  11. 2022Adaptive Trajectory Prediction via Transferable GNNcited by 51
  12. 2020Multi-source Domain Adaptation in the Deep Learning Era: A Systematic Surveycited by 47
  13. 2021Interpretable Social Anchors for Human Trajectory Forecasting in Crowdscited by 44
  14. 2021Towards Robust and Adaptive Motion Forecasting: A Causal Representation Perspectivecited by 40
  15. 2021Vehicle trajectory prediction works, but not everywherecited by 39
  16. 2021Are socially-aware trajectory prediction models really socially-aware?cited by 32
  17. 2022SSL-Lanes: Self-Supervised Learning for Motion Forecasting in Autonomous Drivingcited by 32
  18. 2022Transferable and Adaptable Driving Behavior Predictioncited by 20
  19. 2022Safety-compliant Generative Adversarial Networks for Human Trajectory Forecastingcited by 12

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 4; 2024: 5; 2025: 1; 2026: 2.

12 papers.

  1. 2024UniTraj: A Unified Framework for Scalable Vehicle Trajectory Predictioncited by 52
  2. 2023Visual Tuningcited by 34
  3. 2023Towards Motion Forecasting with Real-World Perception Inputs: Are End-to-End Approaches Competitive?cited by 22
  4. 2025Reinforced Refinement with Self-Aware Expansion for End-to-End Autonomous Drivingcited by 19
  5. 2023Low-rank Adaptation of Large Language Model Rescoring for Parameter-Efficient Speech Recognitioncited by 13
  6. 2023Sim-to-Real Causal Transfer: A Metric Learning Approach to Causally-Aware Interaction Representationscited by 4
  7. 2024Annealed Winner-Takes-All for Motion Forecastingcited by 4
  8. 2024LoRD: Adapting Differentiable Driving Policies to Distribution Shiftscited by 4
  9. 2024Forecast-PEFT: Parameter-Efficient Fine-Tuning for Pre-trained Motion Forecasting Modelscited by 3
  10. 2024PPT: Pretraining with Pseudo-Labeled Trajectories for Motion Forecastingcited by 2
  11. 2026Eye Gaze-Informed and Context-Aware Pedestrian Trajectory Prediction in Shared Spaces with Automated Shuttles: A Virtual Reality Studycited by 1
  12. 2026PRISM: Predictive Representation of Interaction Style and Motion for Social Robot Navigationcited by 0