Image-text Pairs Research Articles

SEMScene: Semantic-Consistency Enhanced Multi-Level Scene Graph Matching for Image-Text Retrieval

Image-text retrieval, a fundamental cross-modal task, performs similarity reasoning for images and texts. The primary challenge for image-text retrieval is cross-modal semantic heterogeneity, where the semantic features of visual and textual modalities are rich but distinct. Scene graph is an effective representation for images and texts as it explicitly models objects and their relations. Existing scene graph based methods have not fully taken the features regarding various granularities implicit in scene graph into consideration (e.g. triplets), the inadequate feature matching incurs the absence of non-trivial semantic information (e.g. inner relations among triplets). Therefore, we propose a S emantic-Consistency E nhanced M ulti-Level Scene Graph Matching (SEMScene) network, which exploits the semantic relevance between visual and textual scene graphs from fine-grained to coarse-grained. Firstly, under the scene graph representation, we perform feature matching including low-level node matching, mid-level semantic triplet matching, and high-level holistic scene graph matching. Secondly, to enhance the semantic-consistency for object-fused triplets carrying key correlation information, we propose a dual-step constraint mechanism in mid-level matching. Thirdly, to guide the model to learn the semantic-consistency of matched image-text pairs, we devise effective loss functions for each stage of the dual-step constraint. Comprehensive experiments on Flickr30K and MS-COCO datasets demonstrate that SEMScene achieves state-of-the-art performances with significant improvements.

A multimodal transfer learning framework for the classification of disaster-related social media images

Rapidly classifying disaster-related social media (SM) images during a catastrophe event is critical for enhancing disaster response efforts. However, the biggest challenge lies in acquiring labeled data for an ongoing (target) disaster to train supervised learning-based models, given that the labeling process is both time-consuming and costly. In this study, we address this challenge by proposing a new multimodal transfer learning framework for the real-time classification of SM images of the target disaster. The proposed framework is based on Contrastive Language-Image Pretraining (CLIP) model, jointly pretrained on a dataset of image-text pairs via contrastive learning. We propose two distinct methods to design our classification framework (1) Zero-Shot CLIP: it learns visual representations from images paired with natural language descriptions of classes. By utilizing the vision and language capabilities of CLIP, we extract meaningful features from unlabeled target disaster images and map them to semantically related textual class descriptions, enabling image classification without training on disaster-specific data. (2) Linear-Probe CLIP: it further enhances the performance and involves training a linear classifier on top of the pretrained CLIP model’s features, specifically tailored to the disaster image classification task. By optimizing the linear-probe classifier, we improve the model’s ability to discriminate between different classes and achieve higher performance without the need for labeled data of the target disaster. Both methods are evaluated on a benchmark X (formerly Twitter) dataset comprising images of seven real-world disaster events. The experimental outcomes showcase the efficacy of the proposed methods, with Linear-Probe CLIP achieving a remarkable 7% improvement in average F1-score relative to the state-of-the-art methods.

Evaluating GPT-V4 (GPT-4 with Vision) on Detection of Radiologic Findings on Chest Radiographs.

Background Generating radiologic findings from chest radiographs is pivotal in medical image analysis. The emergence of OpenAI's generative pretrained transformer, GPT-4 with vision (GPT-4V), has opened new perspectives on the potential for automated image-text pair generation. However, the application of GPT-4V to real-world chest radiography is yet to be thoroughly examined. Purpose To investigate the capability of GPT-4V to generate radiologic findings from real-world chest radiographs. Materials and Methods In this retrospective study, 100 chest radiographs with free-text radiology reports were annotated by a cohort of radiologists, two attending physicians and three residents, to establish a reference standard. Of 100 chest radiographs, 50 were randomly selected from the National Institutes of Health (NIH) chest radiographic data set, and 50 were randomly selected from the Medical Imaging and Data Resource Center (MIDRC). The performance of GPT-4V at detecting imaging findings from each chest radiograph was assessed in the zero-shot setting (where it operates without prior examples) and few-shot setting (where it operates with two examples). Its outcomes were compared with the reference standard with regards to clinical conditions and their corresponding codes in the International Statistical Classification of Diseases, Tenth Revision (ICD-10), including the anatomic location (hereafter, laterality). Results In the zero-shot setting, in the task of detecting ICD-10 codes alone, GPT-4V attained an average positive predictive value (PPV) of 12.3%, average true-positive rate (TPR) of 5.8%, and average F1 score of 7.3% on the NIH data set, and an average PPV of 25.0%, average TPR of 16.8%, and average F1 score of 18.2% on the MIDRC data set. When both the ICD-10 codes and their corresponding laterality were considered, GPT-4V produced an average PPV of 7.8%, average TPR of 3.5%, and average F1 score of 4.5% on the NIH data set, and an average PPV of 10.9%, average TPR of 4.9%, and average F1 score of 6.4% on the MIDRC data set. With few-shot learning, GPT-4V showed improved performance on both data sets. When contrasting zero-shot and few-shot learning, there were improved average TPRs and F1 scores in the few-shot setting, but there was not a substantial increase in the average PPV. Conclusion Although GPT-4V has shown promise in understanding natural images, it had limited effectiveness in interpreting real-world chest radiographs. © RSNA, 2024 Supplemental material is available for this article.

Breaking Through the Noisy Correspondence: A Robust Model for Image-Text Matching

Unleashing the power of image-text matching in real-world applications is hampered by noisy correspondence. Manually curating high-quality datasets is expensive and time-consuming, and datasets generated using diffusion models are not adequately well-aligned. The most promising way is to collect image-text pairs from the Internet, but it will inevitably introduce noisy correspondence. To reduce the negative impact of noisy correspondence, we propose a novel model that first transforms the noisy correspondence filtering problem into a similarity distribution modeling problem by exploiting the powerful capabilities of pre-trained models. Specifically, we use the Gaussian Mixture model to model the similarity obtained by CLIP as clean distribution and noisy distribution, to filter out most of the noisy correspondence in the dataset. Afterward, we used relatively clean data to fine-tune the model. To further reduce the negative impact of unfiltered noisy correspondence, i.e., a minimal part where two distributions intersect during the fine-tuning process, we propose a distribution-sensitive dynamic margin ranking loss, further increasing the distance between the two distributions. Through continuous iteration, the noisy correspondence gradually decreases and the model performance gradually improves. Our extensive experiments demonstrate the effectiveness and robustness of our model even under high noise rates.

A self-supervised framework for cross-modal search in histopathology archives using scale harmonization

The exponential growth of data across various medical domains has generated a substantial demand for techniques to analyze multimodal big data. This demand is particularly pronounced in fields such as computational pathology due to the diverse nature of the tissue. Cross-modal retrieval aims to identify a common latent space where different modalities, such as image-text pairs, exhibit close alignment. The primary challenge, however, often lies in the representation of tissue features. While language models can be trained relatively easily, visual models frequently struggle due to the scarcity of labeled data. To address this issue, the innovative concept of harmonization has been introduced, extending the learning scheme distillation without supervision, known as DINO. The harmonization of scale refines the DINO paradigm through a novel patching approach, overcoming the complexities posed by gigapixel whole slide images in digital pathology. Experiments conducted on diverse datasets have demonstrated that the proposed approach significantly enhances cross-modal retrieval in tissue imaging. Moreover, it exhibits vast potential for other fields that rely on gigapixel imaging.

High-Accuracy Tomato Leaf Disease Image-Text Retrieval Method Utilizing LAFANet.

Tomato leaf disease control in the field of smart agriculture urgently requires attention and reinforcement. This paper proposes a method called LAFANet for image-text retrieval, which integrates image and text information for joint analysis of multimodal data, helping agricultural practitioners to provide more comprehensive and in-depth diagnostic evidence to ensure the quality and yield of tomatoes. First, we focus on six common tomato leaf disease images and text descriptions, creating a Tomato Leaf Disease Image-Text Retrieval Dataset (TLDITRD), introducing image-text retrieval into the field of tomato leaf disease retrieval. Then, utilizing ViT and BERT models, we extract detailed image features and sequences of textual features, incorporating contextual information from image-text pairs. To address errors in image-text retrieval caused by complex backgrounds, we propose Learnable Fusion Attention (LFA) to amplify the fusion of textual and image features, thereby extracting substantial semantic insights from both modalities. To delve further into the semantic connections across various modalities, we propose a False Negative Elimination-Adversarial Negative Selection (FNE-ANS) approach. This method aims to identify adversarial negative instances that specifically target false negatives within the triplet function, thereby imposing constraints on the model. To bolster the model's capacity for generalization and precision, we propose Adversarial Regularization (AR). This approach involves incorporating adversarial perturbations during model training, thereby fortifying its resilience and adaptability to slight variations in input data. Experimental results show that, compared with existing ultramodern models, LAFANet outperformed existing models on TLDITRD dataset, with top1, top5, and top10 reaching 83.3% and 90.0%, and top1, top5, and top10 reaching 80.3%, 93.7%, and 96.3%. LAFANet offers fresh technical backing and algorithmic insights for the retrieval of tomato leaf disease through image-text correlation.

Prompt-Enhanced Generation for Multimodal Open Question Answering

Multimodal open question answering involves retrieving relevant information from both images and their corresponding texts given a question and then generating the answer. The quality of the generated answer heavily depends on the quality of the retrieved image–text pairs. Existing methods encode and retrieve images and texts, inputting the retrieved results into a language model to generate answers. These methods overlook the semantic alignment of image–text pairs within the information source, which affects the encoding and retrieval performance. Furthermore, these methods are highly dependent on retrieval performance, and poor retrieval quality can lead to poor generation performance. To address these issues, we propose a prompt-enhanced generation model, PEG, which includes generating supplementary descriptions for images to provide ample material for image–text alignment while also utilizing vision–language joint encoding to improve encoding effects and thereby enhance retrieval performance. Contrastive learning is used to enhance the model’s ability to discriminate between relevant and irrelevant information sources. Moreover, we further explore the knowledge within pre-trained model parameters through prefix-tuning to generate background knowledge relevant to the questions, offering additional input for answer generation and reducing the model’s dependency on retrieval performance. Experiments conducted on the WebQA and MultimodalQA datasets demonstrate that our model outperforms other baseline models in retrieval and generation performance.

Attr4Vis: Revisiting Importance of Attribute Classification in Vision-Language Models for Video Recognition

Vision-language models (VLMs), pretrained on expansive datasets containing image-text pairs, have exhibited remarkable transferability across a diverse spectrum of visual tasks. The leveraging of knowledge encoded within these potent VLMs holds significant promise for the advancement of effective video recognition models. A fundamental aspect of pretrained VLMs lies in their ability to establish a crucial bridge between the visual and textual domains. In our pioneering work, we introduce the Attr4Vis framework, dedicated to exploring knowledge transfer between Video and Text modalities to bolster video recognition performance. Central to our contributions is the comprehensive revisitation of Text-to-Video classifier initialization, a critical step that refines the initialization process and streamlines the integration of our framework, particularly within existing Vision-Language Models (VLMs). Furthermore, we emphasize the adoption of dense attribute generation techniques, shedding light on their paramount importance in video analysis. By effectively encoding attribute changes over time, these techniques significantly enhance event representation and recognition within videos. In addition, we introduce an innovative Attribute Enrichment Algorithm aimed at enriching set of attributes by large language models (LLMs) like ChatGPT. Through the seamless integration of these components, Attr4Vis attains a state-of-the-art accuracy of 91.5% on the challenging Kinetics-400 dataset using the InternVideo model.

The impact of clinical context on the recognition of facial expressions

IntroductionSeveral authors have demonstrated the relevance of the therapist sensitivity to the affective expression of his client (Merten & Schwab, 2005; 150-158), as well as to his own emotional experience (Haynal-Raymond et al., 2005;142-148) in order to build a more effective therapeutic relationship, and results. An important source of information to decode the emotional expression hints is the face, and its expression (Ekman & Friesen, 1975; Russel & Fernández-Dolls, 1997;275-294). Despite common sense saying that context is relevant to understand the meaning of the emotional facial expression, the literature review shows inconsistent results.ObjectivesThe main goal of this study was to evaluate the impact of clinical context over the perception of the emotional facial expression.MethodsThis study followed a within-subjects design, and its sample consisted of 60 clinical psychologists. 21 combinations of prototypical expression images with mixed emotional signals, and clinical information texts were presented to the participants. Then their judgement on the type of emotion displayed was requested. The presentation of the text-image pairs was randomized between three conditions: consistent, and non-consistent, and neutral.ResultsThe results suggest that emotions are more easily recognized in the presence of a concordant context than a non-concordant or neutral one, and that the greater the similarity between the facial expression of the image presented and the face prototypically associated with the context, the greater the influence of the context.However, In the recognition of mixed emotional signs, there was greater recognition of signs of anger in the facial expression, as a non-dominant emotion, when in the presence of the neutral story than of the story that agreed with the dominant emotion (sadness). There was also greater recognition of sadness, as a non-dominant emotion, in the presence of a story in agreement with fear than in the presence of a neutral story. There was also a statistically significant increase in the attribution of anger to images in which it is not present and whose dominant emotion is fear, when associated with a context of aggression vs. a neutral context.It was also found that there was a significant decrease in the attribution of fear to the sadness-anger image (25%-75%) in the presence of the aggression context compared to the neutral and panic contexts.There was also a statistically significant decrease in the attribution of sadness to an image of fear in the neutral context compared to the other contexts (panic and aggression).ConclusionsIn conclusion, our study have shown an impact of context over overvaluation or the undervaluation of the emotional facial expression as well as either with prototypical expressions or the mixed emotional signals when referring to sadness, fear, and anger. Thus, mental health clinicians should consider the influence of these contexts.Disclosure of InterestNone Declared

Heterogeneous Graph Fusion Network for cross-modal image-text retrieval

Exploring the semantic correspondence of image-text pairs is significant as it bridges vision and language. Most prior works focus on global semantic alignment or local semantic alignment, by developing a fine neural network that facilitates the corresponding alignment but neglects the semantic information and relative position information between image regions, or text words, which will lead to a non-meaningful alignment. To this end, a Heterogeneous Graph Fusion Network (HGFN) is proposed to explore the correlation score of vision-language for improving the accuracy of cross-modal image-text retrieval in this paper. Specifically, we first construct an undirected fully-connected graph based on the semantic or relative position information for each image, as well as a textual graph with neighborhood information of the text. Then, we present a graph fusion module to integrate the features of heterogeneous graphs into a unified hybrid representation, in which the graph convolutional network is utilized to gather neighborhood information to alleviate potentially non-meaningful alignment. In addition, we also propose a novel “Dynamic top-K negative” strategy for the selection of negative examples in the training process. Experimental results demonstrate that HGFN achieves comparable performance with state-of-the-art approaches on the Flickr30K and MSCOCO datasets.

A face retrieval technique combining large models and artificial neural networks

SummaryMultimodal sentiment analysis is a popular research direction in the field of affective computing. It extends unimodal‐based sentiment analysis to the environment based on multimodal information exchange. Face retrieval is the most critical technology in the current multimodal sentiment analysis field. Traditional methods for face retrieval rely on large amounts of data to train the matching relationship between text and face images. This approach suffers from several issues, including high data acquisition costs, insufficient or poor quality training data, and high computational costs for training models. To address the aforementioned issues, this paper proposes a face retrieval technique that merges a large language model with a visual base model. This technique explores the use of zero samples for training, eliminating the need to collect data and train the model to achieve face retrieval. As a result, it significantly reduces the cost requirements of data and computational power. First, an accurate natural text of face description is generated using a large language model based on cue engineering based on face‐independent features. Second, the text and images are encoded using a visual base model and mapped into the same multidimensional space for matching. The experimental results indicate that this method performs well on the CelebA dataset. Extracting 31 different sets of face feature descriptions from the CelebA dataset for retrieval achieves 72% on top1 accuracy and 93% on top3 accuracy. Compared to traditional methods, the method proposed in this paper achieves considerable face retrieval accuracy without collecting text image pairs of data and without training the model.

Image Captioning with Multi-Context Synthetic Data

Image captioning requires numerous annotated image-text pairs, resulting in substantial annotation costs. Recently, large models (e.g. diffusion models and large language models) have excelled in producing high-quality images and text. This potential can be harnessed to create synthetic image-text pairs for training captioning models. Synthetic data can improve cost and time efficiency in data collection, allow for customization to specific domains, bootstrap generalization capability for zero-shot performance, and circumvent privacy concerns associated with real-world data. However, existing methods struggle to attain satisfactory performance solely through synthetic data. We identify the issue as generated images from simple descriptions mostly capture a solitary perspective with limited context, failing to align with the intricate scenes prevalent in real-world imagery. To tackle this, we present an innovative pipeline that introduces multi-context data generation. Beginning with an initial text corpus, our approach employs a large language model to extract multiple sentences portraying the same scene from diverse viewpoints. These sentences are then condensed into a single sentence with multiple contexts. Subsequently, we generate intricate images using the condensed captions through diffusion models. Our model is exclusively trained on synthetic image-text pairs crafted through this process. The effectiveness of our pipeline is validated through experimental results in both the in-domain and cross-domain settings, where it achieves state-of-the-art performance on well-known datasets such as MSCOCO, Flickr30k, and NoCaps.

GroundVLP: Harnessing Zero-Shot Visual Grounding from Vision-Language Pre-training and Open-Vocabulary Object Detection

Visual grounding, a crucial vision-language task involving the understanding of the visual context based on the query expression, necessitates the model to capture the interactions between objects, as well as various spatial and attribute information. However, the annotation data of visual grounding task is limited due to its time-consuming and labor-intensive annotation process, resulting in the trained models being constrained from generalizing its capability to a broader domain. To address this challenge, we propose GroundVLP, a simple yet effective zero-shot method that harnesses visual grounding ability from the existing models trained from image-text pairs and pure object detection data, both of which are more conveniently obtainable and offer a broader domain compared to visual grounding annotation data. GroundVLP proposes a fusion mechanism that combines the heatmap from GradCAM and the object proposals of open-vocabulary detectors. We demonstrate that the proposed method significantly outperforms other zero-shot methods on RefCOCO/+/g datasets, surpassing prior zero-shot state-of-the-art by approximately 28% on the test split of RefCOCO and RefCOCO+. Furthermore, GroundVLP performs comparably to or even better than some non-VLP-based supervised models on the Flickr30k entities dataset. Our code is available at https://github.com/om-ai-lab/GroundVLP.

CLIM: Contrastive Language-Image Mosaic for Region Representation

Detecting objects accurately from a large or open vocabulary necessitates the vision-language alignment on region representations. However, learning such a region-text alignment by obtaining high-quality box annotations with text labels or descriptions is expensive and infeasible. In contrast, collecting image-text pairs is simpler but lacks precise object location information to associate regions with texts. In this paper, we propose a novel approach called Contrastive Language-Image Mosaic (CLIM), which leverages large-scale image-text pairs effectively for aligning region and text representations. CLIM combines multiple images into a mosaicked image and treats each image as a ‘pseudo region’. The feature of each pseudo region is extracted and trained to be similar to the corresponding text embedding while dissimilar from others by a contrastive loss, enabling the model to learn the region-text alignment without costly box annotations. As a generally applicable approach, CLIM consistently improves different open-vocabulary object detection methods that use caption supervision. Furthermore, CLIM can effectively enhance the region representation of vision-language models, thus providing stronger backbones for open-vocabulary object detectors. Our experimental results demonstrate that CLIM improves different baseline open-vocabulary object detectors by a large margin on both OV-COCO and OV-LVIS benchmarks. The code is available at https://github.com/wusize/CLIM.

Expediting Contrastive Language-Image Pretraining via Self-Distilled Encoders

Recent advances in vision language pretraining (VLP) have been largely attributed to the large-scale data collected from the web. However, uncurated dataset contains weakly correlated image-text pairs, causing data inefficiency. To address the issue, knowledge distillation have been explored at the expense of extra image and text momentum encoders to generate teaching signals for misaligned image-text pairs. In this paper, our goal is to resolve the misalignment problem with an efficient distillation framework. To this end, we propose ECLIPSE: Expediting Contrastive Language-Image Pretraining with Self-distilled Encoders. ECLIPSE features a distinctive distillation architecture wherein a shared text encoder is utilized between an online image encoder and a momentum image encoder. This strategic design choice enables the distillation to operate within a unified projected space of text embedding, resulting in better performance. Based on the unified text embedding space, ECLIPSE compensates for the additional computational cost of the momentum image encoder by expediting the online image encoder. Through our extensive experiments, we validate that there is a sweet spot between expedition and distillation where the partial view from the expedited online image encoder interacts complementarily with the momentum teacher. As a result, ECLIPSE outperforms its counterparts while achieving substantial acceleration in inference speed.

Noise-Aware Image Captioning with Progressively Exploring Mismatched Words

Image captioning aims to automatically generate captions for images by learning a cross-modal generator from vision to language. The large amount of image-text pairs required for training is usually sourced from the internet due to the manual cost, which brings the noise with mismatched relevance that affects the learning process. Unlike traditional noisy label learning, the key challenge in processing noisy image-text pairs is to finely identify the mismatched words to make the most use of trustworthy information in the text, rather than coarsely weighing the entire examples. To tackle this challenge, we propose a Noise-aware Image Captioning method (NIC) to adaptively mitigate the erroneous guidance from noise by progressively exploring mismatched words. Specifically, NIC first identifies mismatched words by quantifying word-label reliability from two aspects: 1) inter-modal representativeness, which measures the significance of the current word by assessing cross-modal correlation via prediction certainty; 2) intra-modal informativeness, which amplifies the effect of current prediction by combining the quality of subsequent word generation. During optimization, NIC constructs the pseudo-word-labels considering the reliability of the origin word-labels and model convergence to periodically coordinate mismatched words. As a result, NIC can effectively exploit both clean and noisy image-text pairs to learn a more robust mapping function. Extensive experiments conducted on the MS-COCO and Conceptual Caption datasets validate the effectiveness of our method in various noisy scenarios.

DocMSU: A Comprehensive Benchmark for Document-Level Multimodal Sarcasm Understanding

Multimodal Sarcasm Understanding (MSU) has a wide range of applications in the news field such as public opinion analysis and forgery detection. However, existing MSU benchmarks and approaches usually focus on sentence-level MSU. In document-level news, sarcasm clues are sparse or small and are often concealed in long text. Moreover, compared to sentence-level comments like tweets, which mainly focus on only a few trends or hot topics (e.g., sports events), content in the news is considerably diverse. Models created for sentence-level MSU may fail to capture sarcasm clues in document-level news. To fill this gap, we present a comprehensive benchmark for Document-level Multimodal Sarcasm Understanding (DocMSU). Our dataset contains 102,588 pieces of news with text-image pairs, covering 9 diverse topics such as health, business, etc. The proposed large-scale and diverse DocMSU significantly facilitates the research of document-level MSU in real-world scenarios. To take on the new challenges posed by DocMSU, we introduce a fine-grained sarcasm comprehension method to properly align the pixel-level image features with word-level textual features in documents. Experiments demonstrate the effectiveness of our method, showing that it can serve as a baseline approach to the challenging DocMSU.

SoftCLIP: Softer Cross-Modal Alignment Makes CLIP Stronger

During the preceding biennium, vision-language pre-training has achieved noteworthy success on several downstream tasks. Nevertheless, acquiring high-quality image-text pairs, where the pairs are entirely exclusive of each other, remains a challenging task, and noise exists in the commonly used datasets. To address this issue, we propose SoftCLIP, a novel approach that relaxes the strict one-to-one constraint and achieves a soft cross-modal alignment by introducing a softened target, which is generated from the fine-grained intra-modal self-similarity. The intra-modal guidance is indicative to enable two pairs have some local similarities and model many-to-many relationships between the two modalities. Besides, since the positive still dominates in the softened target distribution, we disentangle the negatives in the distribution to further boost the relation alignment with the negatives in the cross-modal learning. Extensive experiments demonstrate the effectiveness of SoftCLIP. In particular, on ImageNet zero-shot classification task, using CC3M/CC12M as pre-training dataset, SoftCLIP brings a top-1 accuracy improvement of 6.8%/7.2% over the CLIP baseline.

SkyScript: A Large and Semantically Diverse Vision-Language Dataset for Remote Sensing

Remote sensing imagery, despite its broad applications in helping achieve Sustainable Development Goals and tackle climate change, has not yet benefited from the recent advancements of versatile, task-agnostic vision language models (VLMs). A key reason is that the large-scale, semantically diverse image-text dataset required for developing VLMs is still absent for remote sensing images. Unlike natural images, remote sensing images and their associated text descriptions cannot be efficiently collected from the public Internet at scale. In this work, we bridge this gap by using geo-coordinates to automatically connect open, unlabeled remote sensing images with rich semantics covered in OpenStreetMap, and thus construct SkyScript, a comprehensive vision-language dataset for remote sensing images, comprising 2.6 million image-text pairs covering 29K distinct semantic tags. With continual pre-training on this dataset, we obtain a VLM that surpasses baseline models with a 6.2% average accuracy gain in zero-shot scene classification across seven benchmark datasets. It also demonstrates the ability of zero-shot transfer for fine-grained object attribute classification and cross-modal retrieval. We hope this dataset can support the advancement of VLMs for various multi-modal tasks in remote sensing, such as open-vocabulary classification, retrieval, captioning, and text-to-image synthesis.

Improving Cross-Modal Alignment with Synthetic Pairs for Text-Only Image Captioning

Although image captioning models have made significant advancements in recent years, the majority of them heavily depend on high-quality datasets containing paired images and texts which are costly to acquire. Previous works leverage the CLIP's cross-modal association ability for image captioning, relying solely on textual information under unsupervised settings. However, not only does a modality gap exist between CLIP text and image features, but a discrepancy also arises between training and inference due to the unavailability of real-world images, which hinders the cross-modal alignment in text-only captioning. This paper proposes a novel method to address these issues by incorporating synthetic image-text pairs. A pre-trained text-to-image model is deployed to obtain images that correspond to textual data, and the pseudo features of generated images are optimized toward the real ones in the CLIP embedding space. Furthermore, textual information is gathered to represent image features, resulting in the image features with various semantics and the bridged modality gap. To unify training and inference, synthetic image features would serve as the training prefix for the language decoder, while real images are used for inference. Additionally, salient objects in images are detected as assistance to enhance the learning of modality alignment. Experimental results demonstrate that our method obtains the state-of-the-art performance on benchmark datasets.