TRACE: Task-Adaptive Reasoning and Representation Learning for Universal Multimodal Retrieval

TL;DR

TRACE introduces a unified model that combines generative reasoning with discriminative embedding learning for improved multimodal retrieval, enabling better handling of complex queries and zero-shot transfer.

Contribution

It presents TRACE, a novel framework that unifies reasoning and representation learning, and introduces M-BEIR-CoT, a large-scale dataset for training and evaluating such models.

Findings

Achieves state-of-the-art results on M-BEIR benchmark.

Demonstrates implicit routing behavior for complex versus simple queries.

Exhibits strong zero-shot transferability to unseen domains.

Abstract

Universal Multimodal Retrieval requires unified embedding models capable of interpreting diverse user intents, ranging from simple keywords to complex compositional instructions. While Multimodal Large Language Models (MLLMs) possess strong reasoning capabilities, prevailing adaptations confine them to static encoders, underutilizing their generative potential. This encoder-only paradigm struggles with complex intents that demand logical deduction rather than superficial pattern matching. To address this, we introduce TRACE (Task-adaptive Reasoning And Compressing Embeddings). TRACE unifies generative reasoning with discriminative representation learning. It first generates a structured Chain-of-Thought (CoT) to explicitly reason about the query, and subsequently compresses this reasoning trace into a compact embedding via a dedicated token. To train this framework, we construct M-BEIR-CoT, a large-scale dataset featuring a difficulty-aware routing strategy. Experiments on the M-BEIR benchmark establish TRACE as the new state-of-the-art. Crucially, TRACE demonstrates a learned implicit routing behavior. It autonomously activates reasoning for complex queries while bypassing it for simpler ones, achieving an optimal balance between retrieval accuracy and inference throughput. Furthermore, by internalizing the deductive process, TRACE exhibits remarkable zero-shot transferability to unseen domains and novel constraints.