Zero-shot Composed Image Retrieval (ZS-CIR) involves diverse tasks with varied visual manipulation intents across domains, scenes, objects, and attributes. A key challenge is that existing datasets contain limited intent-relevant annotations, making it hard for models to infer human intent from textual modifications. We introduce an intent-centric image–text dataset generated via reasoning by a Multimodal Large Language Model (MLLM) to better train ZS-CIR models for human manipulation intent understanding. Building on this dataset, we propose De-MINDS, a framework that distills the MLLM’s reasoning ability to capture manipulation intent and enhance models’ comprehension of modified text. A simple mapping network translates image information into language space and combines it with the manipulation text to form a query. De-MINDS then extracts intention-relevant information from this query and encodes it as pseudo-word tokens for accurate ZS-CIR. Across four ZS-CIR tasks, De-MINDS shows strong generalization and improves over existing methods by 2.15% to 4.05%, establishing new state-of-the-art results with comparable inference time.

The ability of Large Language Models (LLMs) to use ex ternal tools unlocks powerful real-world interactions, mak ing rigorous evaluation essential. However, current bench marks primarily report final accuracy, revealing what mod els can do but obscuring the cognitive bottlenecks that define their true capability boundaries. To move from simple per formance scoring to a diagnostic tool, we introduce a frame workgroundedinCognitive LoadTheory.Ourframeworkde constructs task complexity into two quantifiable components: Intrinsic Load, the inherent structural complexity of the solu tion path, formalized with a novel Tool Interaction Graph; and Extraneous Load, the difficulty arising from ambiguous task presentation. To enable controlled experiments, we construct ToolLoad-Bench, the first benchmark with parametrically ad justable cognitive load. Our evaluation reveals distinct per formance cliffs as cognitive load increases, allowing us to precisely map each model’s capability boundary. We validate that our framework’s predictions are highly calibrated with empirical results, establishing a principled methodology for understanding an agent’s limits and a practical foundation for building more efficient systems.

Zero-Shot Composed Image Retrieval (ZS-CIR) involves diverse tasks with a broad range of visual content manipulation intent across domain, scene, object, and attribute. The key challenge for ZS-CIR tasks is to modify a reference image according to manipulation text to accurately retrieve a target image, especially when the reference image is missing essential target content. In this paper, we propose a novel prediction-based mapping network, named PrediCIR, to adaptively predict the missing target visual content in reference images in the latent space before mapping for accurate ZS-CIR. Specifically, a world view generation module first constructs a source view by omitting certain visual content of a target view, coupled with an action that includes the manipulation intent derived from existing image-caption pairs. Then, a target content prediction module trains a world model as a predictor to adaptively predict the missing visual information guided by user intention in manipulating text at the latent space. The two modules map an image with the predicted relevant information to a pseudo-word token without extra supervision. Our model shows strong generalization ability on six ZS-CIR tasks. It obtains consistent and significant performance boosts ranging from 1.73% to 4.45% over the best methods and achieves new state-of-the-art results on ZS-CIR. Our code is available at https://github.com/Pter61/predicir.

Composed Image Retrieval (CIR) aims to retrieve target images that closely resemble a reference image while integrating user-specified textual modifications, thereby capturing user intent more accurately. Existing training-free zero-shot CIR (ZS-CIR) methods often employ a two-stage process: they first generate a caption for the reference image and then use Large Language Models for reasoning a target description. However, these methods suffer from missing critical visual details and limited reasoning capabilities, leading to suboptimal retrieval performance. To address these challenges, we propose a novel, training-free one-stage method, One-Stage Reflective Chain-of-Thought Reasoning (OSrCIR) for ZS-CIR, which employs Multimodal Large Language Models to retain essential visual information in a single-stage reasoning process, eliminating the information loss in two-stage methods. Our Reflective Chain-of-Thought framework further improves interpretative accuracy by aligning manipulation intent with contextual cues from reference images. OSrCIR achieves performance gains of 1.80% to 6.44% over existing training-free methods across multiple tasks, setting new state-of-the-art results in ZS-CIR and enhancing its utility in vision-language applications. Our code is available at https://github.com/microsoft/ACV/tree/main/OSrCIR.

Different from the Composed Image Retrieval task that requires expensive labels for training task-specific models, Zero-Shot Composed Image Retrieval (ZS-CIR) involves diverse tasks with a broad range of visual content manipulation intent that could be related to domain, scene, object, and attribute. The key challenge for ZS-CIR tasks is to learn a more accurate image representation that has adaptive attention to the reference image for various manipulation descriptions. In this paper, we propose a novel context-dependent mapping network, named Context-I2W, for adaptively converting description-relevant Image information into a pseudo-word token composed of the description for accurate ZS-CIR. Specifically, an Intent View Selector first dynamically learns a rotation rule to map the identical image to a task-specific manipulation view. Then a Visual Target Extractor further captures local information covering the main targets in ZS-CIR tasks under the guidance of multiple learnable queries. The two complementary modules work together to map an image to a context-dependent pseudo-word token without extra supervision. Our model shows strong generalization ability on four ZS-CIR tasks, including domain conversion, object composition, object manipulation, and attribute manipulation. It obtains consistent and significant performance boosts ranging from 1.88% to 3.60% over the best methods and achieves new state-of-the-art results on ZS-CIR. Our code is available at https://anonymous.4open.science/r/Context-I2W-4224/.

Recent work shows that documents from encyclopedias serve as helpful auxiliary information for zero-shot learning. Existing methods align the entire semantics of a document with corresponding images to transfer knowledge. However, they disregard that semantic information is not equivalent between them, resulting in a suboptimal alignment. In this work, we propose a novel network to extract multi-view semantic concepts from documents and images and align the matching rather than entire concepts. Specifically, we propose a semantic decomposition module to generate multi-view semantic embeddings from visual and textual sides, providing the basic concepts for partial alignment. To alleviate the issue of information redundancy among embeddings, we propose the local-to-semantic variance loss to capture distinct local details and multiple semantic diversity loss to enforce orthogonality among embeddings. Subsequently, two losses are introduced to partially align visual-semantic embedding pairs according to their semantic relevance at the view and word-to-patch levels. Consequently, we consistently outperform state-of-the-art methods under two document sources in three standard benchmarks for document-based zero-shot learning. Qualitatively, we show that our model learns the interpretable partial association. Code is available at https://github.com/MorningStarOvO/EmDepart.

Communication is widely known as the primary bottleneck of federated learning, and quantization of local model updates before uploading to the parameter server is an effective solution to reduce the communication overhead. However, prior literature always assumes homogeneous quantization for all clients, while in reality, devices are heterogeneous and support different levels of quantization precision. This heterogeneity of quantization poses a new challenge: fine-quantized model updates are more accurate than coarse-quantized ones, and how to optimally aggregate them at the server is an open problem. In this paper, we propose FedHQ – Federated Learning with Heterogeneous Quantization – that allocates different aggregation weights to different clients by minimizing the convergence rate upper bound as a function of the heterogeneous quantization errors of all clients, for both strongly convex and non-convex loss functions. To further accelerate the convergence, the instantaneous quantization error is computed and piggybacked when each client uploads the local model update, and the server dynamically calculates the weight accordingly for the current aggregation. Numerical experiment results demonstrate the performance advantages of FedHQ over both vanilla FedAvg with standard equal weights and a heuristic aggregation scheme, which assigns weights linearly proportional to the clients’ quantization precision.