OSCAR: Optimization-Steered Agentic Planning for Composed Image Retrieval

TL;DR

OSCAR introduces a novel optimization-based agentic planning framework for composed image retrieval, replacing heuristic search with principled trajectory optimization to improve accuracy and generalization across benchmarks.

Contribution

It reformulates agentic CIR as a trajectory optimization problem and employs a two-stage mixed-integer programming approach for better planning and retrieval performance.

Findings

Outperforms state-of-the-art methods on multiple benchmarks.

Achieves high performance with only 10% of training data.

Demonstrates strong generalization of planning logic.

Abstract

Composed image retrieval (CIR) requires complex reasoning over heterogeneous visual and textual constraints. Existing approaches largely fall into two paradigms: unified embedding retrieval, which suffers from single-model myopia, and heuristic agentic retrieval, which is limited by suboptimal, trial-and-error orchestration. To this end, we propose OSCAR, an optimization-steered agentic planning framework for composed image retrieval. We are the first to reformulate agentic CIR from a heuristic search process into a principled trajectory optimization problem. Instead of relying on heuristic trial-and-error exploration, OSCAR employs a novel offline-online paradigm. In the offline phase, we model CIR via atomic retrieval selection and composition as a two-stage mixed-integer programming problem, mathematically deriving optimal trajectories that maximize ground-truth coverage for training samples via rigorous boolean set operations. These trajectories are then stored in a golden library to serve as in-context demonstrations for online steering of VLM planner at online inference time. Extensive experiments on three public benchmarks and a private industrial benchmark show that OSCAR consistently outperforms SOTA baselines. Notably, it achieves superior performance using only 10% of training data, demonstrating strong generalization of planning logic rather than dataset-specific memorization.