Large-scale pre-trained text-image models with dual-encoder architectures (such as CLIP) are typically adopted for various vision-language applications, including text-image retrieval. However,these models are still less practical on edge devices or for real-time situations, due to the substantial indexing and inference time and the large consumption of computational resources. Although knowledge distillation techniques have been widely utilized for uni-modal model compression, how to expand them to the situation when the numbers of modalities and teachers/students are doubled has been rarely studied. In this paper, we conduct comprehensive experiments on this topic and propose the fully-Connected knowledge interaction graph (Cona) technique for cross-modal pre-training distillation. Based on our findings, the resulting ConaCLIP achieves SOTA performances on the widely-used Flickr30K and MSCOCO benchmarks under the lightweight setting. An industry application of our method on an e-commercial platform further demonstrates the significant effectiveness of ConaCLIP.
Composed image retrieval aims to find an image that best matches a given multi-modal user query consisting of a reference image and text pair. Existing methods commonly pre-compute image embeddings over the entire corpus and compare these to a reference image embedding modified by the query text at test time. Such a pipeline is very efficient at test time since fast vector distances can be used to evaluate candidates, but modifying the reference image embedding guided only by a short textual description can be difficult, especially independent of potential candidates. An alternative approach is to allow interactions between the query and every possible candidate, i.e., reference-text-candidate triplets, and pick the best from the entire set. Though this approach is more discriminative, for large-scale datasets the computational cost is prohibitive since pre-computation of candidate embeddings is no longer possible. We propose to combine the merits of both schemes using a two-stage model. Our first stage adopts the conventional vector distancing metric and performs a fast pruning among candidates. Meanwhile, our second stage employs a dual-encoder architecture, which effectively attends to the input triplet of reference-text-candidate and re-ranks the candidates. Both stages utilize a vision-and-language pre-trained network, which has proven beneficial for various downstream tasks. Our method consistently outperforms state-of-the-art approaches on standard benchmarks for the task. Our implementation is available at https://github.com/Cuberick-Orion/Candidate-Reranking-CIR.
Perceiving multi-modal information and fulfilling dialogues with humans is a long-term goal of artificial intelligence. Pre-training is commonly regarded as an effective approach for multi-modal dialogue. However, due to the limited availability of multi-modal dialogue data, there is still scarce research on multi-modal dialogue pre-training. Yet another intriguing challenge emerges from the encompassing nature of multi-modal dialogue, which involves various modalities and tasks. Moreover, new forms of tasks may arise at unpredictable points in the future. Hence, it is essential for designed multi-modal dialogue models to possess sufficient flexibility to adapt to such scenarios. This paper proposes \textbf{PaCE}, a unified, structured, compositional multi-modal dialogue pre-training framework. It utilizes a combination of several fundamental experts to accommodate multiple dialogue-related tasks and can be pre-trained using limited dialogue and extensive non-dialogue multi-modal data. Furthermore, we propose a progressive training method where old experts from the past can assist new experts, facilitating the expansion of their capabilities. Experimental results demonstrate that PaCE achieves state-of-the-art results on eight multi-modal dialog benchmarks.
Contrastively trained vision-language models have achieved remarkable progress in vision and language representation learning, leading to state-of-the-art models for various downstream multimodal tasks. However, recent research has highlighted severe limitations of these models in their ability to perform compositional reasoning over objects, attributes, and relations. Scene graphs have emerged as an effective way to understand images compositionally. These are graph-structured semantic representations of images that contain objects, their attributes, and relations with other objects in a scene. In this work, we consider the scene graph parsed from text as a proxy for the image scene graph and propose a graph decomposition and augmentation framework along with a coarse-to-fine contrastive learning objective between images and text that aligns sentences of various complexities to the same image. Along with this, we propose novel negative mining techniques in the scene graph space for improving attribute binding and relation understanding. Through extensive experiments, we demonstrate the effectiveness of our approach that significantly improves attribute binding, relation understanding, systematic generalization, and productivity on multiple recently proposed benchmarks (For example, improvements upto $18\%$ for systematic generalization, $16.5\%$ for relation understanding over a strong baseline), while achieving similar or better performance than CLIP on various general multimodal tasks.
Making image retrieval methods practical for real-world search applications requires significant progress in dataset scales, entity comprehension, and multimodal information fusion. In this work, we introduce \textbf{E}ntity-\textbf{D}riven \textbf{I}mage \textbf{S}earch (EDIS), a challenging dataset for cross-modal image search in the news domain. EDIS consists of 1 million web images from actual search engine results and curated datasets, with each image paired with a textual description. Unlike datasets that assume a small set of single-modality candidates, EDIS reflects real-world web image search scenarios by including a million multimodal image-text pairs as candidates. EDIS encourages the development of retrieval models that simultaneously address cross-modal information fusion and matching. To achieve accurate ranking results, a model must: 1) understand named entities and events from text queries, 2) ground entities onto images or text descriptions, and 3) effectively fuse textual and visual representations. Our experimental results show that EDIS challenges state-of-the-art methods with dense entities and a large-scale candidate set. The ablation study also proves that fusing textual features with visual features is critical in improving retrieval results.
Digital pathology has revolutionized cancer diagnosis by leveraging Content-Based Medical Image Retrieval (CBMIR) for analyzing histopathological Whole Slide Images (WSIs). CBMIR enables searching for similar content, enhancing diagnostic reliability and accuracy. In 2020, breast and prostate cancer constituted 11.7% and 14.1% of cases, respectively, as reported by the Global Cancer Observatory (GCO). The proposed Unsupervised CBMIR (UCBMIR) replicates the traditional cancer diagnosis workflow, offering a dependable method to support pathologists in WSI-based diagnostic conclusions. This approach alleviates pathologists' workload, potentially enhancing diagnostic efficiency. To address the challenge of the lack of labeled histopathological images in CBMIR, a customized unsupervised Convolutional Auto Encoder (CAE) was developed, extracting 200 features per image for the search engine component. UCBMIR was evaluated using widely-used numerical techniques in CBMIR, alongside visual evaluation and comparison with a classifier. The validation involved three distinct datasets, with an external evaluation demonstrating its effectiveness. UCBMIR outperformed previous studies, achieving a top 5 recall of 99% and 80% on BreaKHis and SICAPv2, respectively, using the first evaluation technique. Precision rates of 91% and 70% were achieved for BreaKHis and SICAPv2, respectively, using the second evaluation technique. Furthermore, UCBMIR demonstrated the capability to identify various patterns in patches, achieving an 81% accuracy in the top 5 when tested on an external image from Arvaniti.
Visual text evokes an image in a person's mind, while non-visual text fails to do so. A method to automatically detect visualness in text will enable text-to-image retrieval and generation models to augment text with relevant images. This is particularly challenging with long-form text as text-to-image generation and retrieval models are often triggered for text that is designed to be explicitly visual in nature, whereas long-form text could contain many non-visual sentences. To this end, we curate a dataset of 3,620 English sentences and their visualness scores provided by multiple human annotators. We also propose a fine-tuning strategy that adapts large vision-language models like CLIP by modifying the model's contrastive learning objective to map text identified as non-visual to a common NULL image while matching visual text to their corresponding images in the document. We evaluate the proposed approach on its ability to (i) classify visual and non-visual text accurately, and (ii) attend over words that are identified as visual in psycholinguistic studies. Empirical evaluation indicates that our approach performs better than several heuristics and baseline models for the proposed task. Furthermore, to highlight the importance of modeling the visualness of text, we conduct qualitative analyses of text-to-image generation systems like DALL-E. Project webpage: https://gaurav22verma.github.io/text-visualness/
Word embeddings, i.e., semantically meaningful vector representation of words, are largely influenced by the distributional hypothesis "You shall know a word by the company it keeps" (Harris, 1954), whereas modern prediction-based neural network embeddings rely on design choices and hyperparameter optimization. Word embeddings like Word2Vec, GloVe etc. well capture the contextuality and real-world analogies but contemporary convolution-based image embeddings such as VGGNet, AlexNet, etc. do not capture contextual knowledge. The popular king-queen analogy does not hold true for most commonly used vision embeddings. In this paper, we introduce a pre-trained joint embedding (JE), named IMAGINATOR, trained on 21K distinct image objects level from 1M image+text pairs. JE is a way to encode multimodal data into a vector space where the text modality serves as the ground-ing key, which the complementary modality (in this case, the image) is anchored with. IMAGINATOR encapsulates three individual representations: (i) object-object co-location, (ii) word-object co-location, and (iii) word-object correlation. These three ways capture complementary aspects of the two modalities which are further combined to obtain the final JEs. Generated JEs are intrinsically evaluated to assess how well they capture the contextuality and real-world analogies. We also evaluate pre-trained IMAGINATOR JEs on three downstream tasks: (i) image captioning, (ii) Image2Tweet, and (iii) text-based image retrieval. IMAGINATOR establishes a new standard on the aforementioned down-stream tasks by outperforming the current SoTA on all the selected tasks. IMAGINATOR will be made publicly available. The codes are available at https://github.com/varunakk/IMAGINATOR
Object re-identification (ReID) from images plays a critical role in application domains of image retrieval (surveillance, retail analytics, etc.) and multi-object tracking (autonomous driving, robotics, etc.). However, systems that additionally or exclusively perceive the world from depth sensors are becoming more commonplace without any corresponding methods for object ReID. In this work, we fill the gap by providing the first large-scale study of object ReID from point clouds and establishing its performance relative to image ReID. To enable such a study, we create two large-scale ReID datasets with paired image and LiDAR observations and propose a lightweight matching head that can be concatenated to any set or sequence processing backbone (e.g., PointNet or ViT), creating a family of comparable object ReID networks for both modalities. Run in Siamese style, our proposed point cloud ReID networks can make thousands of pairwise comparisons in real-time ($10$ Hz). Our findings demonstrate that their performance increases with higher sensor resolution and approaches that of image ReID when observations are sufficiently dense. Our strongest network trained at the largest scale achieves ReID accuracy exceeding $90\%$ for rigid objects and $85\%$ for deformable objects (without any explicit skeleton normalization). To our knowledge, we are the first to study object re-identification from real point cloud observations.
Locating a specific mobile application screen from existing repositories is restricted to basic keyword searches, such as Google Image Search, or necessitates a complete query screen image, as in the case of Swire. However, interactive partial sketch-based solutions like PSDoodle have limitations, including inaccuracy and an inability to consider text appearing on the screen. A potentially effective solution involves implementing a system that provides interactive partial sketching functionality for efficiently structuring user interface elements. Additionally, the system should incorporate text queries to enhance its capabilities further. Our approach, TpD, represents the pioneering effort to enable an iterative search of screens by combining interactive sketching and keyword search techniques. TpD is built on a combination of the Rico repository of approximately 58k Android app screens and the PSDoodle. Our evaluation with third-party software developers showed that PSDoodle provided higher top-10 screen retrieval accuracy than state-of-the-art Swire and required less time to complete a query than other interactive solutions.
The remarkable achievements of ChatGPT and GPT-4 have sparked a wave of interest and research in the field of large language models for Artificial General Intelligence (AGI). These models provide intelligent solutions close to human thinking, enabling us to use general artificial intelligence to solve problems in various applications. However, in remote sensing (RS), the scientific literature on the implementation of AGI remains relatively scant. Existing AI-related research in remote sensing primarily focuses on visual understanding tasks while neglecting the semantic understanding of the objects and their relationships. This is where vision-language models excel, as they enable reasoning about images and their associated textual descriptions, allowing for a deeper understanding of the underlying semantics. Vision-language models can go beyond visual recognition of RS images, model semantic relationships, and generate natural language descriptions of the image. This makes them better suited for tasks requiring visual and textual understanding, such as image captioning, and visual question answering. This paper provides a comprehensive review of the research on vision-language models in remote sensing, summarizing the latest progress, highlighting challenges, and identifying potential research opportunities.
We introduce a novel Region-based contrastive pretraining for Medical Image Retrieval (RegionMIR) that demonstrates the feasibility of medical image retrieval with similar anatomical regions. RegionMIR addresses two major challenges for medical image retrieval i) standardization of clinically relevant searching criteria (e.g., anatomical, pathology-based), and ii) localization of anatomical area of interests that are semantically meaningful. In this work, we propose an ROI image retrieval image network that retrieves images with similar anatomy by extracting anatomical features (via bounding boxes) and evaluate similarity between pairwise anatomy-categorized features between the query and the database of images using contrastive learning. ROI queries are encoded using a contrastive-pretrained encoder that was fine-tuned for anatomy classification, which generates an anatomical-specific latent space for region-correlated image retrieval. During retrieval, we compare the anatomically encoded query to find similar features within a feature database generated from training samples, and retrieve images with similar regions from training samples. We evaluate our approach on both anatomy classification and image retrieval tasks using the Chest ImaGenome Dataset. Our proposed strategy yields an improvement over state-of-the-art pretraining and co-training strategies, from 92.24 to 94.12 (2.03%) classification accuracy in anatomies. We qualitatively evaluate the image retrieval performance demonstrating generalizability across multiple anatomies with different morphology.
Zero-shot sketch-based image retrieval (ZS-SBIR) is challenging due to the cross-domain nature of sketches and photos, as well as the semantic gap between seen and unseen image distributions. Previous methods fine-tune pre-trained models with various side information and learning strategies to learn a compact feature space that is shared between the sketch and photo domains and bridges seen and unseen classes. However, these efforts are inadequate in adapting domains and transferring knowledge from seen to unseen classes. In this paper, we present an effective ``Adapt and Align'' approach to address the key challenges. Specifically, we insert simple and lightweight domain adapters to learn new abstract concepts of the sketch domain and improve cross-domain representation capabilities. Inspired by recent advances in image-text foundation models (e.g., CLIP) on zero-shot scenarios, we explicitly align the learned image embedding with a more semantic text embedding to achieve the desired knowledge transfer from seen to unseen classes. Extensive experiments on three benchmark datasets and two popular backbones demonstrate the superiority of our method in terms of retrieval accuracy and flexibility.
In addition to relevance, diversity is an important yet less studied performance metric of cross-modal image retrieval systems, which is critical to user experience. Existing solutions for diversity-aware image retrieval either explicitly post-process the raw retrieval results from standard retrieval systems or try to learn multi-vector representations of images to represent their diverse semantics. However, neither of them is good enough to balance relevance and diversity. On the one hand, standard retrieval systems are usually biased to common semantics and seldom exploit diversity-aware regularization in training, which makes it difficult to promote diversity by post-processing. On the other hand, multi-vector representation methods are not guaranteed to learn robust multiple projections. As a result, irrelevant images and images of rare or unique semantics may be projected inappropriately, which degrades the relevance and diversity of the results generated by some typical algorithms like top-k. To cope with these problems, this paper presents a new method called CoLT that tries to generate much more representative and robust representations for accurately classifying images. Specifically, CoLT first extracts semantics-aware image features by enhancing the preliminary representations of an existing one-to-one cross-modal system with semantics-aware contrastive learning. Then, a transformer-based token classifier is developed to subsume all the features into their corresponding categories. Finally, a post-processing algorithm is designed to retrieve images from each category to form the final retrieval result. Extensive experiments on two real-world datasets Div400 and Div150Cred show that CoLT can effectively boost diversity, and outperforms the existing methods as a whole (with a higher F1 score).
The rise of multi-modal search requests from users has highlighted the importance of multi-modal retrieval (i.e. image-to-text or text-to-image retrieval), yet the more complex task of image-to-multi-modal retrieval, crucial for many industry applications, remains under-explored. To address this gap and promote further research, we introduce and define the concept of Image-to-Multi-Modal Retrieval (IMMR), a process designed to retrieve rich multi-modal (i.e. image and text) documents based on image queries. We focus on representation learning for IMMR and analyze three key challenges for it: 1) skewed data and noisy label in real-world industrial data, 2) the information-inequality between image and text modality of documents when learning representations, 3) effective and efficient training in large-scale industrial contexts. To tackle the above challenges, we propose a novel framework named organizing categories and learning by classification for retrieval (OCLEAR). It consists of three components: 1) a novel category-oriented data governance scheme coupled with a large-scale classification-based learning paradigm, which handles the skewed and noisy data from a data perspective. 2) model architecture specially designed for multi-modal learning, where information-inequality between image and text modality of documents is considered for modality fusion. 3) a hybrid parallel training approach for tackling large-scale training in industrial scenario. The proposed framework achieves SOTA performance on public datasets and has been deployed in a real-world industrial e-commence system, leading to significant business growth. Code will be made publicly available.
Multi-modal search engines have experienced significant growth and widespread use in recent years, making them the second most common internet use. While search engine systems offer a range of services, the image search field has recently become a focal point in the information retrieval community, as the adage goes, "a picture is worth a thousand words". Although popular search engines like Google excel at image search accuracy and agility, there is an ongoing debate over whether their search results can be biased in terms of gender, language, demographics, socio-cultural aspects, and stereotypes. This potential for bias can have a significant impact on individuals' perceptions and influence their perspectives. In this paper, we present our study on bias and fairness in web search, with a focus on keyword-based image search. We first discuss several kinds of biases that exist in search systems and why it is important to mitigate them. We narrow down our study to assessing and mitigating occupational stereotypes in image search, which is a prevalent fairness issue in image retrieval. For the assessment of stereotypes, we take gender as an indicator. We explore various open-source and proprietary APIs for gender identification from images. With these, we examine the extent of gender bias in top-tanked image search results obtained for several occupational keywords. To mitigate the bias, we then propose a fairness-aware re-ranking algorithm that optimizes (a) relevance of the search result with the keyword and (b) fairness w.r.t genders identified. We experiment on 100 top-ranked images obtained for 10 occupational keywords and consider random re-ranking and re-ranking based on relevance as baselines. Our experimental results show that the fairness-aware re-ranking algorithm produces rankings with better fairness scores and competitive relevance scores than the baselines.
The paper proposes a Federated Content-Based Medical Image Retrieval (FedCBMIR) platform that utilizes Federated Learning (FL) to address the challenges of acquiring a diverse medical data set for training CBMIR models. CBMIR assists pathologists in diagnosing breast cancer more rapidly by identifying similar medical images and relevant patches in prior cases compared to traditional cancer detection methods. However, CBMIR in histopathology necessitates a pool of Whole Slide Images (WSIs) to train to extract an optimal embedding vector that leverages search engine performance, which may not be available in all centers. The strict regulations surrounding data sharing in medical data sets also hinder research and model development, making it difficult to collect a rich data set. The proposed FedCBMIR distributes the model to collaborative centers for training without sharing the data set, resulting in shorter training times than local training. FedCBMIR was evaluated in two experiments with three scenarios on BreaKHis and Camelyon17 (CAM17). The study shows that the FedCBMIR method increases the F1-Score (F1S) of each client to 98%, 96%, 94%, and 97% in the BreaKHis experiment with a generalized model of four magnifications and does so in 6.30 hours less time than total local training. FedCBMIR also achieves 98% accuracy with CAM17 in 2.49 hours less training time than local training, demonstrating that our FedCBMIR is both fast and accurate for both pathologists and engineers. In addition, our FedCBMIR provides similar images with higher magnification for non-developed countries where participate in the worldwide FedCBMIR with developed countries to facilitate mitosis measuring in breast cancer diagnosis. We evaluate this scenario by scattering BreaKHis into four centers with different magnifications.
Image retrieval plays an important role in the Internet world. Usually, the core parts of mainstream visual retrieval systems include an online service of the embedding model and a large-scale vector database. For traditional model upgrades, the old model will not be replaced by the new one until the embeddings of all the images in the database are re-computed by the new model, which takes days or weeks for a large amount of data. Recently, backward-compatible training (BCT) enables the new model to be immediately deployed online by making the new embeddings directly comparable to the old ones. For BCT, improving the compatibility of two models with less negative impact on retrieval performance is the key challenge. In this paper, we introduce AdvBCT, an Adversarial Backward-Compatible Training method with an elastic boundary constraint that takes both compatibility and discrimination into consideration. We first employ adversarial learning to minimize the distribution disparity between embeddings of the new model and the old model. Meanwhile, we add an elastic boundary constraint during training to improve compatibility and discrimination efficiently. Extensive experiments on GLDv2, Revisited Oxford (ROxford), and Revisited Paris (RParis) demonstrate that our method outperforms other BCT methods on both compatibility and discrimination. The implementation of AdvBCT will be publicly available at https://github.com/Ashespt/AdvBCT.
Intelligent robots require object-level scene understanding to reason about possible tasks and interactions with the environment. Moreover, many perception tasks such as scene reconstruction, image retrieval, or place recognition can benefit from reasoning on the level of objects. While keypoint-based matching can yield strong results for finding correspondences for images with small to medium view point changes, for large view point changes, matching semantically on the object-level becomes advantageous. In this paper, we propose a learning-based approach which combines local keypoints with novel object-level features for matching object detections between RGB images. We train our object-level matching features based on appearance and inter-frame and cross-frame spatial relations between objects in an associative graph neural network. We demonstrate our approach in a large variety of views on realistically rendered synthetic images. Our approach compares favorably to previous state-of-the-art object-level matching approaches and achieves improved performance over a pure keypoint-based approach for large view-point changes.
Pretrained Vision-Language Models (VLMs) have achieved remarkable performance in image retrieval from text. However, their performance drops drastically when confronted with linguistically complex texts that they struggle to comprehend. Inspired by the Divide-and-Conquer algorithm and dual-process theory, in this paper, we regard linguistically complex texts as compound proposition texts composed of multiple simple proposition sentences and propose an end-to-end Neural Divide-and-Conquer Reasoning framework, dubbed NDCR. It contains three main components: 1) Divide: a proposition generator divides the compound proposition text into simple proposition sentences and produces their corresponding representations, 2) Conquer: a pretrained VLMs-based visual-linguistic interactor achieves the interaction between decomposed proposition sentences and images, 3) Combine: a neural-symbolic reasoner combines the above reasoning states to obtain the final solution via a neural logic reasoning approach. According to the dual-process theory, the visual-linguistic interactor and neural-symbolic reasoner could be regarded as analogical reasoning System 1 and logical reasoning System 2. We conduct extensive experiments on a challenging image retrieval from contextual descriptions data set. Experimental results and analyses indicate NDCR significantly improves performance in the complex image-text reasoning problem. Code link: https://github.com/YunxinLi/NDCR.