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

• URL: http://arxiv.org/abs/2607.00374v1
• Date: Wed, 01 Jul 2026 03:20:06 GMT
• Title: Learning to Compose: Revisiting Proxy Task Design for Zero-Shot Composed Image Retrieval
• Abstract summary: Composed Image Retrieval (CIR) retrieves a target image from a reference image and a textual modification.<n>We propose FoCo, which models composition as two coordinated stages: focusing on modification-relevant visual content, and then completing the target semantics.<n>Experiments on four ZS-CIR benchmarks show FoCo's state-of-the-art performance and improved generalization.
• Score: 31.201448685871558
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: 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.

Related papers

• DiCE-CIR: Direct Composition Learning for Efficient Zero-Shot Composed Image Retrieval [52.15696141238419]
  Zero-shot composed image retrieval (ZS-CIR) aims to retrieve a target image from a multimodal query consisting of a reference image and an edit text.<n>We propose DiCE-CIR, a direct composition learning method that predicts composed query representations by directly composing a reference image and an edit text.
  arXiv  Detail & Related papers  (2026-07-06T04:37:30Z)
• STiTch: Semantic Transition and Transportation in Collaboration for Training-Free Zero-Shot Composed Image Retrieval [38.107904166193364]
  Training-free zero-shot composed image retrieval models are gaining increasing research interest.<n>Recent advances focus on generating the expected target caption.<n>We introduce a novel Semantic Transition and Transportation in collaboration framework for training-free zero-shot CIR tasks.
  arXiv  Detail & Related papers  (2026-05-20T14:51:29Z)
• Constrained Prompt Enhancement for Improving Zero-Shot Generalization of Vision-Language Models [57.357091028792325]
  Vision-language models (VLMs) pre-trained on web-scale data exhibit promising zero-shot generalization but often suffer from semantic misalignment.<n>We propose a novel constrained prompt enhancement (CPE) method to improve visual-textual alignment.<n>Our approach consists of two key components: Topology-Guided Synonymous Semantic Generation (TGSSG) and Category-Agnostic Discriminative Region Selection (CADRS)
  arXiv  Detail & Related papers  (2025-08-24T15:45:22Z)
• Fine-grained Textual Inversion Network for Zero-Shot Composed Image Retrieval [60.20835288280572]
  We propose a novel Fine-grained Textual Inversion Network for ZS-CIR, named FTI4CIR.<n> FTI4CIR comprises two main components: fine-grained pseudo-word token mapping and tri-wise caption-based semantic regularization.
  arXiv  Detail & Related papers  (2025-03-25T02:51:25Z)
• Data-Efficient Generalization for Zero-shot Composed Image Retrieval [67.46975191141928]
  ZS-CIR aims to retrieve the target image based on a reference image and a text description without requiring in-distribution triplets for training.<n>One prevalent approach follows the vision-language pretraining paradigm that employs a mapping network to transfer the image embedding to a pseudo-word token in the text embedding space.<n>We propose a Data-efficient Generalization (DeG) framework, including two novel designs, namely, Textual Supplement (TS) module and Semantic-Set (S-Set)
  arXiv  Detail & Related papers  (2025-03-07T07:49:31Z)
• Spherical Linear Interpolation and Text-Anchoring for Zero-shot Composed Image Retrieval [43.47770490199544]
  Composed Image Retrieval (CIR) is a complex task that retrieves images using a query, which is configured with an image and a caption. We introduce a novel ZS-CIR method that uses Spherical Linear Interpolation (Slerp) to directly merge image and text representations. We also introduce Text-Anchored-Tuning (TAT), a method that fine-tunes the image encoder while keeping the text encoder fixed.
  arXiv  Detail & Related papers  (2024-05-01T15:19:54Z)
• Vision-Aware Text Features in Referring Image Segmentation: From Object Understanding to Context Understanding [26.768147543628096]
  We propose a novel framework that emphasizes object and context comprehension inspired by human cognitive processes. Our method achieves significant performance improvements on three benchmark datasets.
  arXiv  Detail & Related papers  (2024-04-12T16:38:48Z)
• Collaborative Group: Composed Image Retrieval via Consensus Learning from Noisy Annotations [67.92679668612858]
  We propose the Consensus Network (Css-Net), inspired by the psychological concept that groups outperform individuals. Css-Net comprises two core components: (1) a consensus module with four diverse compositors, each generating distinct image-text embeddings; and (2) a Kullback-Leibler divergence loss that encourages learning of inter-compositor interactions. On benchmark datasets, particularly FashionIQ, Css-Net demonstrates marked improvements. Notably, it achieves significant recall gains, with a 2.77% increase in R@10 and 6.67% boost in R@50, underscoring its
  arXiv  Detail & Related papers  (2023-06-03T11:50:44Z)
• BOSS: Bottom-up Cross-modal Semantic Composition with Hybrid Counterfactual Training for Robust Content-based Image Retrieval [61.803481264081036]
  Content-Based Image Retrieval (CIR) aims to search for a target image by concurrently comprehending the composition of an example image and a complementary text. We tackle this task by a novel underlinetextbfBottom-up crunderlinetextbfOss-modal underlinetextbfSemantic compounderlinetextbfSition (textbfBOSS) with Hybrid Counterfactual Training framework.
  arXiv  Detail & Related papers  (2022-07-09T07:14:44Z)
• CRIS: CLIP-Driven Referring Image Segmentation [71.56466057776086]
  We propose an end-to-end CLIP-Driven Referring Image framework (CRIS) CRIS resorts to vision-language decoding and contrastive learning for achieving the text-to-pixel alignment. Our proposed framework significantly outperforms the state-of-the-art performance without any post-processing.
  arXiv  Detail & Related papers  (2021-11-30T07:29:08Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.