Compositional Image Retrieval
PulseAugur coverage of Compositional Image Retrieval — every cluster mentioning Compositional Image Retrieval across labs, papers, and developer communities, ranked by signal.
-
New AMBER framework optimizes vision-language model reranking
Researchers have introduced AMBER, a novel framework for optimizing the use of vision-language models (VLMs) in multimodal retrieval tasks. AMBER addresses the high inference costs associated with VLMs by dynamically al…
-
New framework VMIR-CVI advances zero-shot composed image retrieval
Researchers have introduced VMIR-CVI, a novel framework designed to enhance zero-shot composed image retrieval. This method optimizes multimodal intent representation by converting complex queries into unified textual d…
-
MulVec method enhances zero-shot image retrieval with role-aware matching
Researchers have developed MulVec, a novel method for training-free zero-shot composed image retrieval. Unlike existing approaches that use a single global description, MulVec employs a role-aware system with four disti…
-
New CR-Refiner system enhances 3D scene retrieval with LLM and optimal transport
Researchers have developed CR-Refiner, a novel training-free reranker designed to improve edit-conditioned 3D scene retrieval. This system addresses limitations in existing methods by parsing natural language modificati…
-
Vision-Free CIR Framework Enhances Image Retrieval with LLM Reranking
Researchers have developed a novel vision-free framework for Composed Image Retrieval (CIR), a complex multimodal task. This approach utilizes Attribute-Augmented Hybrid Scoring to compensate for visual details lost in …
-
DiCE-CIR introduces direct composition learning for efficient zero-shot image retrieval
Researchers have introduced DiCE-CIR, a novel direct composition learning method for efficient zero-shot composed image retrieval. This approach bypasses the complex projection and re-encoding steps of previous methods …
-
RankVR framework enhances image retrieval by filtering noisy data
Researchers have introduced RankVR, a new framework designed to improve Composed Image Retrieval (CIR) models. RankVR addresses challenges in large datasets, specifically noisy triplet correspondence, by employing a Glo…
-
New CIRCLED dataset enhances multi-turn image retrieval research
Researchers have introduced CIRCLED, a new multi-turn composed image retrieval dataset designed to overcome the limitations of existing datasets, which often lack dialogue consistency and are confined to specific domain…
-
New framework tackles ambiguity in image retrieval with clarifying questions
Researchers have introduced a new framework for Composed Image Retrieval (CIR) that addresses the inherent ambiguity in queries. Unlike previous systems that assume a single target image, this approach reframes CIR as c…
-
New framework uses multi-agent system for advanced image retrieval
Researchers have introduced a novel framework called PDF for zero-shot compositional image retrieval. This hierarchical multi-agent system aims to overcome limitations in existing methods by incorporating experience sel…