# Move What Matters: Parameter-Efficient Domain Adaptation via Optimal Transport Flow for Collaborative Perception Source: [Move What Matters: Parameter-Efficient Domain Adaptation via Optimal Transport Flow for Collaborative Perception]() Collection index: [llms.txt]() Data updated: 2026-08-05T05:00:18.080Z Authors: Zesheng Jia, Jin Wang, Siao Liu, Lingzhi Li, Ziyao Huang, Yunjiang Xu, Jianping Wang Published: 2026-02-12T04:36:50.000Z ## Abstract Efficient domain adaptation remains a fundamental challenge for deploying multi-agent systems across diverse environments in Vehicle-to-Everything (V2X) collaborative perception. Despite the success of Parameter-Efficient Fine-Tuning (PEFT) in natural language processing and conventional vision tasks, directly applying PEFT to collaborative perception recovers only a limited portion of the performance lost to domain shift. In this work, we identify two complementary bottlenecks that limit this recovery: (i) inter-frame redundancy within a collaborative sequence, which makes the effectiveness of scarce labels sensitive to frame selection, and (ii) foreground cues that become less linearly decodable in deeper-stage representations of a frozen backbone. To address these issues, we propose FlowAdapt, a parameter-efficient framework grounded in optimal transport. Wasserstein Greedy Sampling casts frame selection as minimizing the $W\_{\\infty}$ distance from the sequence to the retained subset, which equals its covering radius, so a farthest first traversal returns a subset provably within twice the optimum. Progressive Knowledge Transfer then routes compressed early-stage features into the deeper stages, gating each stage-local correction by that early evidence. Extensive experiments across target domains and fusion architectures show that FlowAdapt achieves state-of-the-art adaptation performance with about 1\\% trainable parameters, and maintains this lead under localization noise. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Collaborative 3D Object Detection on DAIR-V2X (10% labeled data ratio)](): Collaborative 3D Object Detection; DAIR-V2X 10% labeled data ratio; AP@50: 65.9. - [Collaborative 3D Object Detection on DAIR-V2X (1% subset)](): Collaborative 3D Object Detection; DAIR-V2X 1% labeled data ratio; AP@50: 56.1. - [Collaborative 3D Object Detection on DAIR-V2X (2% labeled data ratio)](): Collaborative 3D Object Detection; DAIR-V2X 2% labeled data ratio; AP@50: 59.1. - [Collaborative 3D Object Detection on DAIR-V2X (5% subset)](): Collaborative 3D Object Detection; DAIR-V2X 5% labeled data ratio; AP@50: 63.1. - [Collaborative 3D Object Detection on DAIR-V2X (20% labeled data ratio)](): Collaborative 3D Object Detection; DAIR-V2X 20% labeled data ratio; AP@50: 67.7. - [3D Object Detection on DAIR-V2X adapted from OPV2V](): 3D Object Detection; DAIR-V2X adapted from OPV2V; AP@50: 63.8. - [3D Object Detection on V2XSet adapted from OPV2V](): 3D Object Detection; V2XSet adapted from OPV2V; AP@IoU=50%: 93.4. ## Original sources and implementation [Original paper]() [PDF]()