The emergence of large Vision Language Models (VLMs) has broadened the scope and capabilities of single-modal Large Language Models (LLMs) by integrating visual modalities, thereby unlocking transformative cross-modal applications in a variety of real-world scenarios. Despite their impressive performance, VLMs are prone to significant hallucinations, particularly in the form of cross-modal inconsistencies. Building on the success of Reinforcement Learning from Human Feedback (RLHF) in aligning LLMs, recent advancements have focused on applying direct preference optimization (DPO) on carefully curated datasets to mitigate these issues. Yet, such approaches typically introduce preference signals in a brute-force manner, neglecting the crucial role of visual information in the alignment process. In this paper, we introduce Re-Align, a novel alignment framework that leverages image retrieval to construct a dual-preference dataset, effectively incorporating both textual and visual preference signals. We further introduce rDPO, an extension of the standard direct preference optimization that incorporates an additional visual preference objective during fine-tuning. Our experimental results demonstrate that Re-Align not only mitigates hallucinations more effectively than previous methods but also yields significant performance gains in general visual question-answering (VQA) tasks. Moreover, we show that Re-Align maintains robustness and scalability across a wide range of VLM sizes and architectures. This work represents a significant step forward in aligning multimodal LLMs, paving the way for more reliable and effective cross-modal applications. We release all the code in https://github.com/taco-group/Re-Align.
This paper explores the possibility of learning custom tokens for representing new concepts in Vision-Language Models (VLMs). Our aim is to learn tokens that can be effective for both discriminative and generative tasks while composing well with words to form new input queries. The targeted concept is specified in terms of a small set of images and a parent concept described using text. We operate on CLIP text features and propose to use a combination of a textual inversion loss and a classification loss to ensure that text features of the learned token are aligned with image features of the concept in the CLIP embedding space. We restrict the learned token to a low-dimensional subspace spanned by tokens for attributes that are appropriate for the given super-class. These modifications improve the quality of compositions of the learned token with natural language for generating new scenes. Further, we show that learned custom tokens can be used to form queries for text-to-image retrieval task, and also have the important benefit that composite queries can be visualized to ensure that the desired concept is faithfully encoded. Based on this, we introduce the method of Generation Aided Image Retrieval, where the query is modified at inference time to better suit the search intent. On the DeepFashion2 dataset, our method improves Mean Reciprocal Retrieval (MRR) over relevant baselines by 7%.
This work introduces ILIAS, a new test dataset for Instance-Level Image retrieval At Scale. It is designed to evaluate the ability of current and future foundation models and retrieval techniques to recognize particular objects. The key benefits over existing datasets include large scale, domain diversity, accurate ground truth, and a performance that is far from saturated. ILIAS includes query and positive images for 1,000 object instances, manually collected to capture challenging conditions and diverse domains. Large-scale retrieval is conducted against 100 million distractor images from YFCC100M. To avoid false negatives without extra annotation effort, we include only query objects confirmed to have emerged after 2014, i.e. the compilation date of YFCC100M. An extensive benchmarking is performed with the following observations: i) models fine-tuned on specific domains, such as landmarks or products, excel in that domain but fail on ILIAS ii) learning a linear adaptation layer using multi-domain class supervision results in performance improvements, especially for vision-language models iii) local descriptors in retrieval re-ranking are still a key ingredient, especially in the presence of severe background clutter iv) the text-to-image performance of the vision-language foundation models is surprisingly close to the corresponding image-to-image case. website: https://vrg.fel.cvut.cz/ilias/
Measuring perceptual similarity is a key tool in computer vision. In recent years perceptual metrics based on features extracted from neural networks with large and diverse training sets, e.g. CLIP, have become popular. At the same time, the metrics extracted from features of neural networks are not adversarially robust. In this paper we show that adversarially robust CLIP models, called R-CLIP$_\textrm{F}$, obtained by unsupervised adversarial fine-tuning induce a better and adversarially robust perceptual metric that outperforms existing metrics in a zero-shot setting, and further matches the performance of state-of-the-art metrics while being robust after fine-tuning. Moreover, our perceptual metric achieves strong performance on related tasks such as robust image-to-image retrieval, which becomes especially relevant when applied to "Not Safe for Work" (NSFW) content detection and dataset filtering. While standard perceptual metrics can be easily attacked by a small perturbation completely degrading NSFW detection, our robust perceptual metric maintains high accuracy under an attack while having similar performance for unperturbed images. Finally, perceptual metrics induced by robust CLIP models have higher interpretability: feature inversion can show which images are considered similar, while text inversion can find what images are associated to a given prompt. This also allows us to visualize the very rich visual concepts learned by a CLIP model, including memorized persons, paintings and complex queries.
Image retrieval has been employed as a robust complementary technique to address the challenge of Unmanned Aerial Vehicles (UAVs) self-positioning. However, most existing methods primarily focus on localizing objects captured by UAVs through complex part-based representations, often overlooking the unique challenges associated with UAV self-positioning, such as fine-grained spatial discrimination requirements and dynamic scene variations. To address the above issues, we propose the Context-Enhanced method for precise UAV Self-Positioning (CEUSP), specifically designed for UAV self-positioning tasks. CEUSP integrates a Dynamic Sampling Strategy (DSS) to efficiently select optimal negative samples, while the Rubik's Cube Attention (RCA) module, combined with the Context-Aware Channel Integration (CACI) module, enhances feature representation and discrimination by exploiting interdimensional interactions, inspired by the rotational mechanics of a Rubik's Cube. Extensive experimental validate the effectiveness of the proposed method, demonstrating notable improvements in feature representation and UAV self-positioning accuracy within complex urban environments. Our approach achieves state-of-the-art performance on the DenseUAV dataset, which is specifically designed for dense urban contexts, and also delivers competitive results on the widely recognized University-1652 benchmark.
Multi-modal models, such as CLIP, have demonstrated strong performance in aligning visual and textual representations, excelling in tasks like image retrieval and zero-shot classification. Despite this success, the mechanisms by which these models utilize training data, particularly the role of memorization, remain unclear. In uni-modal models, both supervised and self-supervised, memorization has been shown to be essential for generalization. However, it is not well understood how these findings would apply to CLIP, which incorporates elements from both supervised learning via captions that provide a supervisory signal similar to labels, and from self-supervised learning via the contrastive objective. To bridge this gap in understanding, we propose a formal definition of memorization in CLIP (CLIPMem) and use it to quantify memorization in CLIP models. Our results indicate that CLIP's memorization behavior falls between the supervised and self-supervised paradigms, with "mis-captioned" samples exhibiting highest levels of memorization. Additionally, we find that the text encoder contributes more to memorization than the image encoder, suggesting that mitigation strategies should focus on the text domain. Building on these insights, we propose multiple strategies to reduce memorization while at the same time improving utility--something that had not been shown before for traditional learning paradigms where reducing memorization typically results in utility decrease.
We address the challenge of generating fair and unbiased image retrieval results given neutral textual queries (with no explicit gender or race connotations), while maintaining the utility (performance) of the underlying vision-language (VL) model. Previous methods aim to disentangle learned representations of images and text queries from gender and racial characteristics. However, we show these are inadequate at alleviating bias for the desired equal representation result, as there usually exists test-time bias in the target retrieval set. So motivated, we introduce a straightforward technique, Post-hoc Bias Mitigation (PBM), that post-processes the outputs from the pre-trained vision-language model. We evaluate our algorithm on real-world image search datasets, Occupation 1 and 2, as well as two large-scale image-text datasets, MS-COCO and Flickr30k. Our approach achieves the lowest bias, compared with various existing bias-mitigation methods, in text-based image retrieval result while maintaining satisfactory retrieval performance. The source code is publicly available at \url{https://anonymous.4open.science/r/Fair_Text_based_Image_Retrieval-D8B2}.
Diffusion models enable high-quality and diverse visual content synthesis. However, they struggle to generate rare or unseen concepts. To address this challenge, we explore the usage of Retrieval-Augmented Generation (RAG) with image generation models. We propose ImageRAG, a method that dynamically retrieves relevant images based on a given text prompt, and uses them as context to guide the generation process. Prior approaches that used retrieved images to improve generation, trained models specifically for retrieval-based generation. In contrast, ImageRAG leverages the capabilities of existing image conditioning models, and does not require RAG-specific training. Our approach is highly adaptable and can be applied across different model types, showing significant improvement in generating rare and fine-grained concepts using different base models. Our project page is available at: https://rotem-shalev.github.io/ImageRAG
Non-native speakers with limited vocabulary often struggle to name specific objects despite being able to visualize them, e.g., people outside Australia searching for numbats. Further, users may want to search for such elusive objects with difficult-to-sketch interactions, e.g., numbat digging in the ground. In such common but complex situations, users desire a search interface that accepts composite multimodal queries comprising hand-drawn sketches of difficult-to-name but easy-to-draw objects and text describing difficult-to-sketch but easy-to-verbalize object attributes or interaction with the scene. This novel problem statement distinctly differs from the previously well-researched TBIR (text-based image retrieval) and SBIR (sketch-based image retrieval) problems. To study this under-explored task, we curate a dataset, CSTBIR (Composite Sketch+Text Based Image Retrieval), consisting of approx. 2M queries and 108K natural scene images. Further, as a solution to this problem, we propose a pretrained multimodal transformer-based baseline, STNET (Sketch+Text Network), that uses a hand-drawn sketch to localize relevant objects in the natural scene image, and encodes the text and image to perform image retrieval. In addition to contrastive learning, we propose multiple training objectives that improve the performance of our model. Extensive experiments show that our proposed method outperforms several state-of-the-art retrieval methods for text-only, sketch-only, and composite query modalities. We make the dataset and code available at our project website.
With the rapid development of AI-generated content (AIGC), the creation of high-quality AI-generated videos has become faster and easier, resulting in the Internet being flooded with all kinds of video content. However, the impact of these videos on the content ecosystem remains largely unexplored. Video information retrieval remains a fundamental approach for accessing video content. Building on the observation that retrieval models often favor AI-generated content in ad-hoc and image retrieval tasks, we investigate whether similar biases emerge in the context of challenging video retrieval, where temporal and visual factors may further influence model behavior. To explore this, we first construct a comprehensive benchmark dataset containing both real and AI-generated videos, along with a set of fair and rigorous metrics to assess bias. This benchmark consists of 13,000 videos generated by two state-of-the-art open-source video generation models. We meticulously design a suite of rigorous metrics to accurately measure this preference, accounting for potential biases arising from the limited frame rate and suboptimal quality of AIGC videos. We then applied three off-the-shelf video retrieval models to perform retrieval tasks on this hybrid dataset. Our findings reveal a clear preference for AI-generated videos in retrieval. Further investigation shows that incorporating AI-generated videos into the training set of retrieval models exacerbates this bias. Unlike the preference observed in image modalities, we find that video retrieval bias arises from both unseen visual and temporal information, making the root causes of video bias a complex interplay of these two factors. To mitigate this bias, we fine-tune the retrieval models using a contrastive learning approach. The results of this study highlight the potential implications of AI-generated videos on retrieval systems.
Zero-shot composed image retrieval (ZS-CIR) enables image search using a reference image and text prompt without requiring specialized text-image composition networks trained on large-scale paired data. However, current ZS-CIR approaches face three critical limitations in their reliance on composed text embeddings: static query embedding representations, insufficient utilization of image embeddings, and suboptimal performance when fusing text and image embeddings. To address these challenges, we introduce the Prompt Directional Vector (PDV), a simple yet effective training-free enhancement that captures semantic modifications induced by user prompts. PDV enables three key improvements: (1) dynamic composed text embeddings where prompt adjustments are controllable via a scaling factor, (2) composed image embeddings through semantic transfer from text prompts to image features, and (3) weighted fusion of composed text and image embeddings that enhances retrieval by balancing visual and semantic similarity. Our approach serves as a plug-and-play enhancement for existing ZS-CIR methods with minimal computational overhead. Extensive experiments across multiple benchmarks demonstrate that PDV consistently improves retrieval performance when integrated with state-of-the-art ZS-CIR approaches, particularly for methods that generate accurate compositional embeddings. The code will be publicly available.
Astronauts take thousands of photos of Earth per day from the International Space Station, which, once localized on Earth's surface, are used for a multitude of tasks, ranging from climate change research to disaster management. The localization process, which has been performed manually for decades, has recently been approached through image retrieval solutions: given an astronaut photo, find its most similar match among a large database of geo-tagged satellite images, in a task called Astronaut Photography Localization (APL). Yet, existing APL approaches are trained only using satellite images, without taking advantage of the millions open-source astronaut photos. In this work we present the first APL pipeline capable of leveraging astronaut photos for training. We first produce full localization information for 300,000 manually weakly labeled astronaut photos through an automated pipeline, and then use these images to train a model, called AstroLoc. AstroLoc learns a robust representation of Earth's surface features through two losses: astronaut photos paired with their matching satellite counterparts in a pairwise loss, and a second loss on clusters of satellite imagery weighted by their relevance to astronaut photography via unsupervised mining. We find that AstroLoc achieves a staggering 35% average improvement in recall@1 over previous SOTA, pushing the limits of existing datasets with a recall@100 consistently over 99%. Finally, we note that AstroLoc, without any fine-tuning, provides excellent results for related tasks like the lost-in-space satellite problem and historical space imagery localization.
In AI-facilitated teaching, leveraging various query styles to interpret abstract text descriptions is crucial for ensuring high-quality teaching. However, current retrieval models primarily focus on natural text-image retrieval, making them insufficiently tailored to educational scenarios due to the ambiguities in the retrieval process. In this paper, we propose a diverse expression retrieval task tailored to educational scenarios, supporting retrieval based on multiple query styles and expressions. We introduce the STEM Education Retrieval Dataset (SER), which contains over 24,000 query pairs of different styles, and the Uni-Retrieval, an efficient and style-diversified retrieval vision-language model based on prompt tuning. Uni-Retrieval extracts query style features as prototypes and builds a continuously updated Prompt Bank containing prompt tokens for diverse queries. This bank can updated during test time to represent domain-specific knowledge for different subject retrieval scenarios. Our framework demonstrates scalability and robustness by dynamically retrieving prompt tokens based on prototype similarity, effectively facilitating learning for unknown queries. Experimental results indicate that Uni-Retrieval outperforms existing retrieval models in most retrieval tasks. This advancement provides a scalable and precise solution for diverse educational needs.
Decoding visual images from brain activity has significant potential for advancing brain-computer interaction and enhancing the understanding of human perception. Recent approaches align the representation spaces of images and brain activity to enable visual decoding. In this paper, we introduce the use of human-aligned image encoders to map brain signals to images. We hypothesize that these models more effectively capture perceptual attributes associated with the rapid visual stimuli presentations commonly used in visual brain data recording experiments. Our empirical results support this hypothesis, demonstrating that this simple modification improves image retrieval accuracy by up to 21% compared to state-of-the-art methods. Comprehensive experiments confirm consistent performance improvements across diverse EEG architectures, image encoders, alignment methods, participants, and brain imaging modalities.
Pre-trained multi-modal Vision-Language Models like CLIP are widely used off-the-shelf for a variety of applications. In this paper, we show that the common practice of individually exploiting the text or image encoders of these powerful multi-modal models is highly suboptimal for intra-modal tasks like image-to-image retrieval. We argue that this is inherently due to the CLIP-style inter-modal contrastive loss that does not enforce any intra-modal constraints, leading to what we call intra-modal misalignment. To demonstrate this, we leverage two optimization-based modality inversion techniques that map representations from their input modality to the complementary one without any need for auxiliary data or additional trained adapters. We empirically show that, in the intra-modal tasks of image-to-image and text-to-text retrieval, approaching these tasks inter-modally significantly improves performance with respect to intra-modal baselines on more than fifteen datasets. Additionally, we demonstrate that approaching a native inter-modal task (e.g. zero-shot image classification) intra-modally decreases performance, further validating our findings. Finally, we show that incorporating an intra-modal term in the pre-training objective or narrowing the modality gap between the text and image feature embedding spaces helps reduce the intra-modal misalignment. The code is publicly available at: https://github.com/miccunifi/Cross-the-Gap.
In this paper, we expand the domain of sketch research into the field of image segmentation, aiming to establish freehand sketches as a query modality for subjective image segmentation. Our innovative approach introduces a "sketch-in-the-loop" image segmentation framework, enabling the segmentation of visual concepts partially, completely, or in groupings - a truly "freestyle" approach - without the need for a purpose-made dataset (i.e., mask-free). This framework capitalises on the synergy between sketch-based image retrieval (SBIR) models and large-scale pre-trained models (CLIP or DINOv2). The former provides an effective training signal, while fine-tuned versions of the latter execute the subjective segmentation. Additionally, our purpose-made augmentation strategy enhances the versatility of our sketch-guided mask generation, allowing segmentation at multiple granularity levels. Extensive evaluations across diverse benchmark datasets underscore the superior performance of our method in comparison to existing approaches across various evaluation scenarios.
Interactive Text-to-Image Retrieval (I-TIR) has emerged as a transformative user-interactive tool for applications in domains such as e-commerce and education. Yet, current methodologies predominantly depend on finetuned Multimodal Large Language Models (MLLMs), which face two critical limitations: (1) Finetuning imposes prohibitive computational overhead and long-term maintenance costs. (2) Finetuning narrows the pretrained knowledge distribution of MLLMs, reducing their adaptability to novel scenarios. These issues are exacerbated by the inherently dynamic nature of real-world I-TIR systems, where queries and image databases evolve in complexity and diversity, often deviating from static training distributions. To overcome these constraints, we propose Diffusion Augmented Retrieval (DAR), a paradigm-shifting framework that bypasses MLLM finetuning entirely. DAR synergizes Large Language Model (LLM)-guided query refinement with Diffusion Model (DM)-based visual synthesis to create contextually enriched intermediate representations. This dual-modality approach deciphers nuanced user intent more holistically, enabling precise alignment between textual queries and visually relevant images. Rigorous evaluations across four benchmarks reveal DAR's dual strengths: (1) Matches state-of-the-art finetuned I-TIR models on straightforward queries without task-specific training. (2) Scalable Generalization: Surpasses finetuned baselines by 7.61% in Hits@10 (top-10 accuracy) under multi-turn conversational complexity, demonstrating robustness to intricate, distributionally shifted interactions. By eliminating finetuning dependencies and leveraging generative-augmented representations, DAR establishes a new trajectory for efficient, adaptive, and scalable cross-modal retrieval systems.
Composed Image Retrieval (CIR) provides an effective way to manage and access large-scale visual data. Construction of the CIR model utilizes triplets that consist of a reference image, modification text describing desired changes, and a target image that reflects these changes. For effectively training CIR models, extensive manual annotation to construct high-quality training datasets, which can be time-consuming and labor-intensive, is required. To deal with this problem, this paper proposes a novel triplet synthesis method by leveraging counterfactual image generation. By controlling visual feature modifications via counterfactual image generation, our approach automatically generates diverse training triplets without any manual intervention. This approach facilitates the creation of larger and more expressive datasets, leading to the improvement of CIR model's performance.
Multi-modal models require aligned, shared embedding spaces. However, common CLIP-based approaches need large amounts of samples and do not natively support 3D or tabular data, both of which are crucial in the medical domain. To address these issues, we revisit CLIP-style alignment by training a domain-specific 3D foundation model as an image encoder and demonstrate that modality alignment is feasible with only 62 MRI scans. Our approach is enabled by a simple embedding accumulation strategy required for training in 3D, which scales the amount of negative pairs across batches in order to stabilize training. We perform a thorough evaluation of various design choices, including the choice of backbone and loss functions, and evaluate the proposed methodology on zero-shot classification and image-retrieval tasks. While zero-shot image-retrieval remains challenging, zero-shot classification results demonstrate that the proposed approach can meaningfully align the representations of 3D MRI with tabular data.
The existence of noisy labels in real-world data negatively impacts the performance of deep learning models. Although much research effort has been devoted to improving the robustness towards noisy labels in classification tasks, the problem of noisy labels in deep metric learning (DML) remains under-explored. Existing noisy label learning methods designed for DML mainly discard suspicious noisy samples, resulting in a waste of the training data. To address this issue, we propose a noise-robust DML framework with SubGroup-based Positive-pair Selection (SGPS), which constructs reliable positive pairs for noisy samples to enhance the sample utilization. Specifically, SGPS first effectively identifies clean and noisy samples by a probability-based clean sample selectionstrategy. To further utilize the remaining noisy samples, we discover their potential similar samples based on the subgroup information given by a subgroup generation module and then aggregate them into informative positive prototypes for each noisy sample via a positive prototype generation module. Afterward, a new contrastive loss is tailored for the noisy samples with their selected positive pairs. SGPS can be easily integrated into the training process of existing pair-wise DML tasks, like image retrieval and face recognition. Extensive experiments on multiple synthetic and real-world large-scale label noise datasets demonstrate the effectiveness of our proposed method. Without any bells and whistles, our SGPS framework outperforms the state-of-the-art noisy label DML methods. Code is available at \url{https://github.com/smuelpeng/SGPS-NoiseFreeDML}.