TRACE-Bench：多參考影像生成之分解與診斷

摘要

儘管近年來統一多模態模型在多參考圖像生成方面取得了進展，現有的基準仍圍繞預定義任務類型（例如「主題組合」）進行組織，而這種方式並不適合此類組合設定，導致覆蓋範圍零散、複雜度不可控且診斷價值有限。鑒於多樣化的多參考任務共享一組共同的原子操作，我們採取以能力為導向的視角，並形式化定義四個算子：錨定（Anchor, f）、解耦（Disentangle, g）、應用（Apply, ⊕）與組合（Compose, C）。任何多參考提示詞皆可表示為這些算子上的組合公式，其結構複雜度以算子槽位數量進行量化。基於此形式化框架，我們建構了 TRACE-Bench，包含約 1,600 個評估案例，槽位數涵蓋 1 至 8，由 631 個公式模板與約 4,000 張涵蓋多種藝術風格及真實世界主體的參考圖像所構成。公式結構直接驅動了算子對齊的評估協議，以進行按能力評分，並透過診斷樹分析實現遞迴失敗定位。對 9 個領先模型的評估揭示出整體評分無法察覺的洞見：主要瓶頸在於解耦（g）與屬性綁定（⊕），而非場景級組合（C）；即使是最佳模型，其屬性保真度得分也僅為 0.74。專案頁面：https://amuseum-whr.github.io/TraceBench

English

Despite recent advances in unified multimodal models for multi-reference image generation, existing benchmarks remain organized around predefined task types (e.g., "subject composition"), which are ill-suited to this combinatorial setting and lead to fragmented coverage, uncontrolled complexity, and little diagnostic value. Recognizing that diverse multi-reference tasks share a common set of atomic operations, we adopt a capability-oriented perspective and formalize four operators: Anchor (f), Disentangle (g), Apply (oplus), and Compose (C). Any multi-reference prompt can then be represented as a compositional formula over these operators, whose structural complexity is quantified by the number of operator slots. Building on this formulation, we construct TRACE-Bench, comprising approximately 1,600 evaluation cases across slot counts 1--8, built from 631 formula templates and around 4,000 reference images spanning diverse artistic styles and real-world subjects. The formula structure directly drives an operator-aligned evaluation protocol for per-capability scoring and a diagnostic tree analysis for recursive failure localization. Evaluating 9 leading models reveals insights invisible to holistic scoring: the primary bottleneck lies in disentanglement (g) and attribute binding (oplus) rather than scene-level composition (C), with even the best model scoring only 0.74 on attribute fidelity. Project page: https://amuseum-whr.github.io/TraceBench