Research map / Multi-agent systems and games / Human-AI and goal-conditioned

One is Plenty: A Polymorphic Feature Interpreter for Immutable Heterogeneous Collaborative Perception

CVPR 2025 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Collaborative multi-agent perception · 5 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes PolyInter, a polymorphic feature interpreter that adapts to new heterogeneous neighbor agents by fine-tuning only per-agent learnable prompts while keeping a shared interpreter network frozen, for immutable heterogeneous collaborative perception.

Problem

Immutable heterogeneous collaborative perception in autonomous driving, where agents have fixed, different perception networks, creating a semantic gap in exchanged intermediate features; existing interpreters need retraining per new agent type or cause cumulative loss in two-stage standard-space in

Approach

A single interpreter network (channel selection module + spatial attention module) with a shared general prompt and per-neighbor specific prompts; new agents are integrated by overriding only their specific prompt (with low-rank decomposition), enabling one-stage interpretation into the ego semantic space.

Key results

• PolyInter improves AP by 7.9% at IoU 0.5 and 11.1% at IoU 0.7 over SOTA interpreters PnPDA and MPDA on OPV2V.
• Adapting to new agents trains only about 1.4% (0.16M) of the sd1 interpreter parameters while still surpassing PnPDA and MPDA.
• Removing the specific prompt drops AP drastically (e.g., 23.3/7.8 on pp8-pp4 vs 77.2/65.8 full).

Measured effects reported in the paper

Surprising findings

• Removing the specific prompt causes a severe performance collapse, showing the prompt is essential rather than a minor add-on.

Limitations

• Evaluation reported mainly on OPV2V in the main text; V2XSet and DAIR-V2X only in supplementary material.
• Performance remains lower for some heterogeneous pairs (e.g., pp8-vn6 AP@0.5 lower than pp8-pp4 baseline numbers).

Experimental setup

Datasets: OPV2V

Benchmarks: OPV2V

Metrics: AP@0.5 AP@0.7

Baselines: PnPDA MPDA

Builds on: MPDA PnPDA visual prompt tuning CoBEVT F-cooper

Closest papers by meaning

• An Extensible Framework for Open Heterogeneous Collaborative Perception (ICLR 2024 · similarity 0.91)
• STAMP: Scalable Task- And Model-agnostic Collaborative Perception (ICLR 2025 · similarity 0.89)
• Pragmatic Heterogeneous Collaborative Perception via Generative Communication Mechanism (NeurIPS 2025 · similarity 0.88)
• NegoCollab: A Common Representation Negotiation Approach for Heterogeneous Collaborative Perception (NeurIPS 2025 · similarity 0.88)
• CoPEFT: Fast Adaptation Framework for Multi-Agent Collaborative Perception with Parameter-Efficient Fine-Tuning (AAAI 2025 · similarity 0.87)
• Privacy-Preserving V2X Collaborative Perception Integrating Unknown Collaborators (AAAI 2025 · similarity 0.87)