# Learning to Compose: Revisiting Proxy Task Design for Zero-Shot Composed Image Retrieval

arXiv: https://arxiv.org/abs/2607.00374 · HTML: https://mlsanity.com/paper/2607.00374/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 24 of its 37 references that are in the map.
- First posted on arXiv 1 Jul 2026.

## Builds on (most cited first, 24 of 24)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation](https://mlsanity.com/paper/2201.12086/) (2022)
- [Sigmoid Loss for Language Image Pre-Training](https://mlsanity.com/paper/2303.15343/) (2023)
- [Image Retrieval on Real-life Images with Pre-trained Vision-and-Language Models](https://mlsanity.com/paper/2108.04024/) (2021)
- [Composing Text and Image for Image Retrieval - An Empirical Odyssey](https://mlsanity.com/paper/1812.07119/) (2018)
- [Zero-Shot Composed Image Retrieval with Textual Inversion](https://mlsanity.com/paper/2303.15247/) (2023)
- [Pic2Word: Mapping Pictures to Words for Zero-shot Composed Image Retrieval](https://mlsanity.com/paper/2302.03084/) (2023)
- [ARTEMIS: Attention-based Retrieval with Text-Explicit Matching and Implicit Similarity](https://mlsanity.com/paper/2203.08101/) (2022)
- [DreamLIP: Language-Image Pre-training with Long Captions](https://mlsanity.com/paper/2403.17007/) (2024)
- [Sentence-level Prompts Benefit Composed Image Retrieval](https://mlsanity.com/paper/2310.05473/) (2023)
- [Context-I2W: Mapping Images to Context-dependent Words for Accurate Zero-Shot Composed Image Retrieval](https://mlsanity.com/paper/2309.16137/) (2023)
- [Modality-Agnostic Attention Fusion for visual search with text feedback](https://mlsanity.com/paper/2007.00145/) (2020)
- [GeneCIS: A Benchmark for General Conditional Image Similarity](https://mlsanity.com/paper/2306.07969/) (2023)
- [Spherical Linear Interpolation and Text-Anchoring for Zero-shot Composed Image Retrieval](https://mlsanity.com/paper/2405.00571/) (2024)
- [Target-Guided Composed Image Retrieval](https://mlsanity.com/paper/2309.01366/) (2023)
- [Missing Target-Relevant Information Prediction with World Model for Accurate Zero-Shot Composed Image Retrieval](https://mlsanity.com/paper/2503.17109/) (2025)
- [Pretrain like Your Inference: Masked Tuning Improves Zero-Shot Composed Image Retrieval](https://mlsanity.com/paper/2311.07622/) (2023)
- [Unveiling the Compositional Ability Gap in Vision-Language Reasoning Model](https://mlsanity.com/paper/2505.19406/) (2025)
- [Leveraging Large Vision-Language Model as User Intent-aware Encoder for Composed Image Retrieval](https://mlsanity.com/paper/2412.11087/) (2024)
- [Zero-shot Composed Image Retrieval Considering Query-target Relationship Leveraging Masked Image-text Pairs](https://mlsanity.com/paper/2406.18836/) (2024)
- [Prompting Large Vision-Language Models for Compositional Reasoning](https://mlsanity.com/paper/2401.11337/) (2024)
- [TMCIR: Token Merge Benefits Composed Image Retrieval](https://mlsanity.com/paper/2504.10995/) (2025)
- [DetailFusion: A Dual-branch Framework with Detail Enhancement for Composed Image Retrieval](https://mlsanity.com/paper/2505.17796/) (2025)

## Built on by (most cited first, 1 of 1)

- [Preserve-and-Compose Training for Composed Image Retrieval](https://mlsanity.com/paper/2609.31202/) (2026)

## Abstract (arXiv)

Composed Image Retrieval (CIR) retrieves a target image from a reference image and a textual modification. While supervised CIR relies on costly triplets, Zero-Shot CIR (ZS-CIR) alleviates this reliance through proxy tasks trained on image-text pairs. However, existing proxy tasks primarily enhance visual and textual representations to accommodate a predefined composition mechanism such as pseudo-word injection into a frozen text encoder or linear feature arithmetic. As a result, the composition function itself remains unlearned, limiting the model's ability to express diverse and fine-grained semantic modifications. To address this, we propose FoCo, which models composition as two coordinated stages: focusing on modification-relevant visual content, and then completing the target semantics. We realize these through two proxy tasks: text-anchored visual aggregation to selectively gather visual content guided by localized textual semantics, and context-conditioned semantic completion to transform these aggregated visuals with the remaining scene context into a coherent composed representation. The tasks are trained jointly with a cross-instance contrastive objective, encouraging semantic diversity and discouraging shortcut composition strategies. Extensive experiments on four ZS-CIR benchmarks show FoCo's state-of-the-art performance and improved generalization.