With the increasing amount of multimedia data on modern mobile systems and IoT infrastructures, harnessing these rich multimodal data without breaching user privacy becomes a critical issue. Federated learning (FL) serves as a privacy-conscious alternative to centralized machine learning. However, existing FL methods extended to multimodal data all rely on model aggregation on single modality level, which restrains the server and clients to have identical model architecture for each modality. This limits the global model in terms of both model complexity and data capacity, not to mention task diversity. In this work, we propose Contrastive Representation Ensemble and Aggregation for Multimodal FL (CreamFL), a multimodal federated learning framework that enables training larger server models from clients with heterogeneous model architectures and data modalities, while only communicating knowledge on public dataset. To achieve better multimodal representation fusion, we design a global-local cross-modal ensemble strategy to aggregate client representations. To mitigate local model drift caused by two unprecedented heterogeneous factors stemming from multimodal discrepancy (modality gap and task gap), we further propose two inter-modal and intra-modal contrasts to regularize local training, which complements information of the absent modality for uni-modal clients and regularizes local clients to head towards global consensus. Thorough evaluations and ablation studies on image-text retrieval and visual question answering tasks showcase the superiority of CreamFL over state-of-the-art FL methods and its practical value.
Often, deep network models are purely inductive during training and while performing inference on unseen data. Thus, when such models are used for predictions, it is well known that they often fail to capture the semantic information and implicit dependencies that exist among objects (or concepts) on a population level. Moreover, it is still unclear how domain or prior modal knowledge can be specified in a backpropagation friendly manner, especially in large-scale and noisy settings. In this work, we propose an end-to-end vision and language model incorporating explicit knowledge graphs. We also introduce an interactive out-of-distribution (OOD) layer using implicit network operator. The layer is used to filter noise that is brought by external knowledge base. In practice, we apply our model on several vision and language downstream tasks including visual question answering, visual reasoning, and image-text retrieval on different datasets. Our experiments show that it is possible to design models that perform similarly to state-of-art results but with significantly fewer samples and training time.
Large-scale vision-language pre-trained models have shown promising transferability to various downstream tasks. As the size of these foundation models and the number of downstream tasks grow, the standard full fine-tuning paradigm becomes unsustainable due to heavy computational and storage costs. This paper proposes UniAdapter, which unifies unimodal and multimodal adapters for parameter-efficient cross-modal adaptation on pre-trained vision-language models. Specifically, adapters are distributed to different modalities and their interactions, with the total number of tunable parameters reduced by partial weight sharing. The unified and knowledge-sharing design enables powerful cross-modal representations that can benefit various downstream tasks, requiring only 1.0%-2.0% tunable parameters of the pre-trained model. Extensive experiments on 6 cross-modal downstream benchmarks (including video-text retrieval, image-text retrieval, VideoQA, and VQA) show that in most cases, UniAdapter not only outperforms the state-of-the-arts, but even beats the full fine-tuning strategy. Particularly, on the MSRVTT retrieval task, UniAdapter achieves 49.7% recall@1 with 2.2% model parameters, outperforming the latest competitors by 2.0%. The code and models are available at https://github.com/RERV/UniAdapter.
Image-text retrieval (ITR) is a task to retrieve the relevant images/texts, given the query from another modality. The conventional dense retrieval paradigm relies on encoding images and texts into dense representations using dual-stream encoders, however, it faces challenges with low retrieval speed in large-scale retrieval scenarios. In this work, we propose the lexicon-weighting paradigm, where sparse representations in vocabulary space are learned for images and texts to take advantage of the bag-of-words models and efficient inverted indexes, resulting in significantly reduced retrieval latency. A crucial gap arises from the continuous nature of image data, and the requirement for a sparse vocabulary space representation. To bridge this gap, we introduce a novel pre-training framework, Lexicon-Bottlenecked Language-Image Pre-Training (LexLIP), that learns importance-aware lexicon representations. This framework features lexicon-bottlenecked modules between the dual-stream encoders and weakened text decoders, allowing for constructing continuous bag-of-words bottlenecks to learn lexicon-importance distributions. Upon pre-training with same-scale data, our LexLIP achieves state-of-the-art performance on two benchmark ITR datasets, MSCOCO and Flickr30k. Furthermore, in large-scale retrieval scenarios, LexLIP outperforms CLIP with a 5.5 ~ 221.3X faster retrieval speed and 13.2 ~ 48.8X less index storage memory.
Real-world data contains a vast amount of multimodal information, among which vision and language are the two most representative modalities. Moreover, increasingly heavier models, \textit{e}.\textit{g}., Transformers, have attracted the attention of researchers to model compression. However, how to compress multimodal models, especially vison-language Transformers, is still under-explored. This paper proposes the \textbf{U}nified and \textbf{P}r\textbf{o}gressive \textbf{P}runing (\textbf{\emph{UPop}}) as a universal vison-language Transformer compression framework, which incorporates 1) unifiedly searching multimodal subnets in a continuous optimization space from the original model, which enables automatic assignment of pruning ratios among compressible modalities and structures; 2) progressively searching and retraining the subnet, which maintains convergence between the search and retrain to attain higher compression ratios. Experiments on various tasks, datasets, and model architectures demonstrate the effectiveness and versatility of the proposed UPop framework. The code is available at https://github.com/sdc17/UPop.
Image and text retrieval is one of the foundational tasks in the vision and language domain with multiple real-world applications. State-of-the-art approaches, e.g. CLIP, ALIGN, represent images and texts as dense embeddings and calculate the similarity in the dense embedding space as the matching score. On the other hand, sparse semantic features like bag-of-words models are more interpretable, but believed to suffer from inferior accuracy than dense representations. In this work, we show that it is possible to build a sparse semantic representation that is as powerful as, or even better than, dense presentations. We extend the CLIP model and build a sparse text and image representation (STAIR), where the image and text are mapped to a sparse token space. Each token in the space is a (sub-)word in the vocabulary, which is not only interpretable but also easy to integrate with existing information retrieval systems. STAIR model significantly outperforms a CLIP model with +$4.9\%$ and +$4.3\%$ absolute Recall@1 improvement on COCO-5k text$\rightarrow$image and image$\rightarrow$text retrieval respectively. It also achieved better performance on both of ImageNet zero-shot and linear probing compared to CLIP.
Vision-language alignment learning for video-text retrieval arouses a lot of attention in recent years. Most of the existing methods either transfer the knowledge of image-text pretraining model to video-text retrieval task without fully exploring the multi-modal information of videos, or simply fuse multi-modal features in a brute force manner without explicit guidance. In this paper, we integrate multi-modal information in an explicit manner by tagging, and use the tags as the anchors for better video-text alignment. Various pretrained experts are utilized for extracting the information of multiple modalities, including object, person, motion, audio, etc. To take full advantage of these information, we propose the TABLE (TAgging Before aLignmEnt) network, which consists of a visual encoder, a tag encoder, a text encoder, and a tag-guiding cross-modal encoder for jointly encoding multi-frame visual features and multi-modal tags information. Furthermore, to strengthen the interaction between video and text, we build a joint cross-modal encoder with the triplet input of [vision, tag, text] and perform two additional supervised tasks, Video Text Matching (VTM) and Masked Language Modeling (MLM). Extensive experimental results demonstrate that the TABLE model is capable of achieving State-Of-The-Art (SOTA) performance on various video-text retrieval benchmarks, including MSR-VTT, MSVD, LSMDC and DiDeMo.
Image-text pretrained models, e.g., CLIP, have shown impressive general multi-modal knowledge learned from large-scale image-text data pairs, thus attracting increasing attention for their potential to improve visual representation learning in the video domain. In this paper, based on the CLIP model, we revisit temporal modeling in the context of image-to-video knowledge transferring, which is the key point for extending image-text pretrained models to the video domain. We find that current temporal modeling mechanisms are tailored to either high-level semantic-dominant tasks (e.g., retrieval) or low-level visual pattern-dominant tasks (e.g., recognition), and fail to work on the two cases simultaneously. The key difficulty lies in modeling temporal dependency while taking advantage of both high-level and low-level knowledge in CLIP model. To tackle this problem, we present Spatial-Temporal Auxiliary Network (STAN) -- a simple and effective temporal modeling mechanism extending CLIP model to diverse video tasks. Specifically, to realize both low-level and high-level knowledge transferring, STAN adopts a branch structure with decomposed spatial-temporal modules that enable multi-level CLIP features to be spatial-temporally contextualized. We evaluate our method on two representative video tasks: Video-Text Retrieval and Video Recognition. Extensive experiments demonstrate the superiority of our model over the state-of-the-art methods on various datasets, including MSR-VTT, DiDeMo, LSMDC, MSVD, Kinetics-400, and Something-Something-V2. Codes will be available at https://github.com/farewellthree/STAN
State-of-the-art video-text retrieval (VTR) methods typically involve fully fine-tuning a pre-trained model (e.g. CLIP) on specific datasets. However, this can result in significant storage costs in practical applications as a separate model per task must be stored. To address this issue, we present our pioneering work that enables parameter-efficient VTR using a pre-trained model, with only a small number of tunable parameters during training. Towards this goal, we propose a new method dubbed Multimodal Video Adapter (MV-Adapter) for efficiently transferring the knowledge in the pre-trained CLIP from image-text to video-text. Specifically, MV-Adapter utilizes bottleneck structures in both video and text branches, along with two novel components. The first is a Temporal Adaptation Module that is incorporated in the video branch to introduce global and local temporal contexts. We also train weights calibrations to adjust to dynamic variations across frames. The second is Cross Modality Tying that generates weights for video/text branches through sharing cross modality factors, for better aligning between modalities. Thanks to above innovations, MV-Adapter can achieve comparable or better performance than standard full fine-tuning with negligible parameters overhead. Notably, MV-Adapter consistently outperforms various competing methods in V2T/T2V tasks with large margins on five widely used VTR benchmarks (MSR-VTT, MSVD, LSMDC, DiDemo, and ActivityNet).
As a fundamental and challenging task in bridging language and vision domains, Image-Text Retrieval (ITR) aims at searching for the target instances that are semantically relevant to the given query from the other modality, and its key challenge is to measure the semantic similarity across different modalities. Although significant progress has been achieved, existing approaches typically suffer from two major limitations: (1) It hurts the accuracy of the representation by directly exploiting the bottom-up attention based region-level features where each region is equally treated. (2) It limits the scale of negative sample pairs by employing the mini-batch based end-to-end training mechanism. To address these limitations, we propose a Unified Semantic Enhancement Momentum Contrastive Learning (USER) method for ITR. Specifically, we delicately design two simple but effective Global representation based Semantic Enhancement (GSE) modules. One learns the global representation via the self-attention algorithm, noted as Self-Guided Enhancement (SGE) module. The other module benefits from the pre-trained CLIP module, which provides a novel scheme to exploit and transfer the knowledge from an off-the-shelf model, noted as CLIP-Guided Enhancement (CGE) module. Moreover, we incorporate the training mechanism of MoCo into ITR, in which two dynamic queues are employed to enrich and enlarge the scale of negative sample pairs. Meanwhile, a Unified Training Objective (UTO) is developed to learn from mini-batch based and dynamic queue based samples. Extensive experiments on the benchmark MSCOCO and Flickr30K datasets demonstrate the superiority of both retrieval accuracy and inference efficiency. Our source code will be released at https://github.com/zhangy0822/USER.
Many models have been proposed for vision and language tasks, especially the image-text retrieval task. All state-of-the-art (SOTA) models in this challenge contained hundreds of millions of parameters. They also were pretrained on a large external dataset that has been proven to make a big improvement in overall performance. It is not easy to propose a new model with a novel architecture and intensively train it on a massive dataset with many GPUs to surpass many SOTA models, which are already available to use on the Internet. In this paper, we proposed a compact graph-based framework, named HADA, which can combine pretrained models to produce a better result, rather than building from scratch. First, we created a graph structure in which the nodes were the features extracted from the pretrained models and the edges connecting them. The graph structure was employed to capture and fuse the information from every pretrained model with each other. Then a graph neural network was applied to update the connection between the nodes to get the representative embedding vector for an image and text. Finally, we used the cosine similarity to match images with their relevant texts and vice versa to ensure a low inference time. Our experiments showed that, although HADA contained a tiny number of trainable parameters, it could increase baseline performance by more than 3.6% in terms of evaluation metrics in the Flickr30k dataset. Additionally, the proposed model did not train on any external dataset and did not require many GPUs but only 1 to train due to its small number of parameters. The source code is available at https://github.com/m2man/HADA.
Legal text retrieval serves as a key component in a wide range of legal text processing tasks such as legal question answering, legal case entailment, and statute law retrieval. The performance of legal text retrieval depends, to a large extent, on the representation of text, both query and legal documents. Based on good representations, a legal text retrieval model can effectively match the query to its relevant documents. Because legal documents often contain long articles and only some parts are relevant to queries, it is quite a challenge for existing models to represent such documents. In this paper, we study the use of attentive neural network-based text representation for statute law document retrieval. We propose a general approach using deep neural networks with attention mechanisms. Based on it, we develop two hierarchical architectures with sparse attention to represent long sentences and articles, and we name them Attentive CNN and Paraformer. The methods are evaluated on datasets of different sizes and characteristics in English, Japanese, and Vietnamese. Experimental results show that: i) Attentive neural methods substantially outperform non-neural methods in terms of retrieval performance across datasets and languages; ii) Pretrained transformer-based models achieve better accuracy on small datasets at the cost of high computational complexity while lighter weight Attentive CNN achieves better accuracy on large datasets; and iii) Our proposed Paraformer outperforms state-of-the-art methods on COLIEE dataset, achieving the highest recall and F2 scores in the top-N retrieval task.
Large-scale vision-language models such as CLIP have shown impressive performance on zero-shot image classification and image-to-text retrieval. However, such performance does not realize in tasks that require a finer-grained correspondence between vision and language, such as Visual Question Answering (VQA). As a potential cause of the difficulty of applying these models to VQA and similar tasks, we report an interesting phenomenon of vision-language models, which we call the Concept Association Bias (CAB). We find that models with CAB tend to treat input as a bag of concepts and attempt to fill in the other missing concept crossmodally, leading to an unexpected zero-shot prediction. We demonstrate CAB by showing that CLIP's zero-shot classification performance greatly suffers when there is a strong concept association between an object (e.g. eggplant) and an attribute (e.g. color purple). We also show that the strength of CAB predicts the performance on VQA. We observe that CAB is prevalent in vision-language models trained with contrastive losses, even when autoregressive losses are jointly employed. However, a model that solely relies on autoregressive loss seems to exhibit minimal or no signs of CAB.
There is increasing adoption of artificial intelligence in drug discovery. However, existing studies use machine learning to mainly utilize the chemical structures of molecules but ignore the vast textual knowledge available in chemistry. Incorporating textual knowledge enables us to realize new drug design objectives, adapt to text-based instructions and predict complex biological activities. Here we present a multi-modal molecule structure-text model, MoleculeSTM, by jointly learning molecules' chemical structures and textual descriptions via a contrastive learning strategy. To train MoleculeSTM, we construct a large multi-modal dataset, namely, PubChemSTM, with over 280,000 chemical structure-text pairs. To demonstrate the effectiveness and utility of MoleculeSTM, we design two challenging zero-shot tasks based on text instructions, including structure-text retrieval and molecule editing. MoleculeSTM has two main properties: open vocabulary and compositionality via natural language. In experiments, MoleculeSTM obtains the state-of-the-art generalization ability to novel biochemical concepts across various benchmarks.
Dense retrievers have made significant strides in text retrieval and open-domain question answering. However, most of these achievements have relied heavily on extensive human-annotated supervision. In this study, we aim to develop unsupervised methods for improving dense retrieval models. We propose two approaches that enable annotation-free and scalable training by creating pseudo querydocument pairs: query extraction and transferred query generation. The query extraction method involves selecting salient spans from the original document to generate pseudo queries. On the other hand, the transferred query generation method utilizes generation models trained for other NLP tasks, such as summarization, to produce pseudo queries. Through extensive experimentation, we demonstrate that models trained using these augmentation methods can achieve comparable, if not better, performance than multiple strong dense baselines. Moreover, combining these strategies leads to further improvements, resulting in superior performance of unsupervised dense retrieval, unsupervised domain adaptation and supervised finetuning, benchmarked on both BEIR and ODQA datasets. Code and datasets are publicly available at https://github.com/salesforce/AugTriever.
Recent years have witnessed the rapid progress of image captioning. However, the demands for large memory storage and heavy computational burden prevent these captioning models from being deployed on mobile devices. The main obstacles lie in the heavyweight visual feature extractors (i.e., object detectors) and complicated cross-modal fusion networks. To this end, we propose LightCap, a lightweight image captioner for resource-limited devices. The core design is built on the recent CLIP model for efficient image captioning. To be specific, on the one hand, we leverage the CLIP model to extract the compact grid features without relying on the time-consuming object detectors. On the other hand, we transfer the image-text retrieval design of CLIP to image captioning scenarios by devising a novel visual concept extractor and a cross-modal modulator. We further optimize the cross-modal fusion model and parallel prediction heads via sequential and ensemble distillations. With the carefully designed architecture, our model merely contains 40M parameters, saving the model size by more than 75% and the FLOPs by more than 98% in comparison with the current state-of-the-art methods. In spite of the low capacity, our model still exhibits state-of-the-art performance on prevalent datasets, e.g., 136.6 CIDEr on COCO Karpathy test split. Testing on the smartphone with only a single CPU, the proposed LightCap exhibits a fast inference speed of 188ms per image, which is ready for practical applications.
Image-text retrieval (ITR) is a challenging task in the field of multimodal information processing due to the semantic gap between different modalities. In recent years, researchers have made great progress in exploring the accurate alignment between image and text. However, existing works mainly focus on the fine-grained alignment between image regions and sentence fragments, which ignores the guiding significance of context background information. Actually, integrating the local fine-grained information and global context background information can provide more semantic clues for retrieval. In this paper, we propose a novel Hierarchical Graph Alignment Network (HGAN) for image-text retrieval. First, to capture the comprehensive multimodal features, we construct the feature graphs for the image and text modality respectively. Then, a multi-granularity shared space is established with a designed Multi-granularity Feature Aggregation and Rearrangement (MFAR) module, which enhances the semantic corresponding relations between the local and global information, and obtains more accurate feature representations for the image and text modalities. Finally, the ultimate image and text features are further refined through three-level similarity functions to achieve the hierarchical alignment. To justify the proposed model, we perform extensive experiments on MS-COCO and Flickr30K datasets. Experimental results show that the proposed HGAN outperforms the state-of-the-art methods on both datasets, which demonstrates the effectiveness and superiority of our model.
Multimodal image-text models have shown remarkable performance in the past few years. However, evaluating robustness against distribution shifts is crucial before adopting them in real-world applications. In this work, we investigate the robustness of 12 popular open-sourced image-text models under common perturbations on five tasks (image-text retrieval, visual reasoning, visual entailment, image captioning, and text-to-image generation). In particular, we propose several new multimodal robustness benchmarks by applying 17 image perturbation and 16 text perturbation techniques on top of existing datasets. We observe that multimodal models are not robust to image and text perturbations, especially to image perturbations. Among the tested perturbation methods, character-level perturbations constitute the most severe distribution shift for text, and zoom blur is the most severe shift for image data. We also introduce two new robustness metrics (\textbf{MMI} for MultiModal Impact score and \textbf{MOR} for Missing Object Rate) for proper evaluations of multimodal models. We hope our extensive study sheds light on new directions for the development of robust multimodal models. More details can be found on the project webpage: \url{https://MMRobustness.github.io}.
Vision Transformers convert images to sequences by slicing them into patches. The size of these patches controls a speed/accuracy tradeoff, with smaller patches leading to higher accuracy at greater computational cost, but changing the patch size typically requires retraining the model. In this paper, we demonstrate that simply randomizing the patch size at training time leads to a single set of weights that performs well across a wide range of patch sizes, making it possible to tailor the model to different compute budgets at deployment time. We extensively evaluate the resulting model, which we call FlexiViT, on a wide range of tasks, including classification, image-text retrieval, open-world detection, panoptic segmentation, and semantic segmentation, concluding that it usually matches, and sometimes outperforms, standard ViT models trained at a single patch size in an otherwise identical setup. Hence, FlexiViT training is a simple drop-in improvement for ViT that makes it easy to add compute-adaptive capabilities to most models relying on a ViT backbone architecture. Code and pre-trained models are available at https://github.com/google-research/big_vision
Disentangled representation learning remains challenging as the underlying factors of variation in the data do not naturally exist. The inherent complexity of real-world data makes it unfeasible to exhaustively enumerate and encapsulate all its variations within a finite set of factors. However, it is worth noting that most real-world data have linguistic equivalents, typically in the form of textual descriptions. These linguistic counterparts can represent the data and effortlessly decomposed into distinct tokens. In light of this, we present Vocabulary Disentangled Retrieval (VDR), a retrieval-based framework that harnesses natural language as proxies of the underlying data variation to drive disentangled representation learning. Our approach employ a bi-encoder model to represent both data and natural language in a vocabulary space, enabling the model to distinguish dimensions that capture intrinsic characteristics within data through its natural language counterpart, thus facilitating disentanglement. We extensively assess the performance of VDR across 15 retrieval benchmark datasets, covering text-to-text and cross-modal retrieval scenarios, as well as human evaluation. Our experimental results compellingly demonstrate the superiority of VDR over previous bi-encoder retrievers with comparable model size and training costs, achieving an impressive 8.7% improvement in NDCG@10 on the BEIR benchmark, a 5.3% increase on MS COCO, and a 6.0% increase on Flickr30k in terms of mean recall in the zero-shot setting. Moreover, The results from human evaluation indicate that interpretability of our method is on par with SOTA captioning models.