Latent text representations exhibit geometric regularities, such as the famous analogy: queen is to king what woman is to man. Such structured semantic relations were not demonstrated on image representations. Recent works aiming at bridging this semantic gap embed images and text into a multimodal space, enabling the transfer of text-defined transformations to the image modality. We introduce the SIMAT dataset to evaluate the task of Image Retrieval with Multimodal queries. SIMAT contains 6k images and 18k textual transformation queries that aim at either replacing scene elements or changing pairwise relationships between scene elements. The goal is to retrieve an image consistent with the (source image, text transformation) query. We use an image/text matching oracle (OSCAR) to assess whether the image transformation is successful. The SIMAT dataset will be publicly available. We use SIMAT to evaluate the geometric properties of multimodal embedding spaces trained with an image/text matching objective, like CLIP. We show that vanilla CLIP embeddings are not very well suited to transform images with delta vectors, but that a simple finetuning on the COCO dataset can bring dramatic improvements. We also study whether it is beneficial to leverage pretrained universal sentence encoders (FastText, LASER and LaBSE).
Recent progress has shown that large-scale pre-training using contrastive image-text pairs can be a promising alternative for high-quality visual representation learning from natural language supervision. Benefiting from a broader source of supervision, this new paradigm exhibits impressive transferability to downstream classification tasks and datasets. However, the problem of transferring the knowledge learned from image-text pairs to more complex dense prediction tasks has barely been visited. In this work, we present a new framework for dense prediction by implicitly and explicitly leveraging the pre-trained knowledge from CLIP. Specifically, we convert the original image-text matching problem in CLIP to a pixel-text matching problem and use the pixel-text score maps to guide the learning of dense prediction models. By further using the contextual information from the image to prompt the language model, we are able to facilitate our model to better exploit the pre-trained knowledge. Our method is model-agnostic, which can be applied to arbitrary dense prediction systems and various pre-trained visual backbones including both CLIP models and ImageNet pre-trained models. Extensive experiments demonstrate the superior performance of our methods on semantic segmentation, object detection, and instance segmentation tasks. Code is available at https://github.com/raoyongming/DenseCLIP
Recently, by introducing large-scale dataset and strong transformer network, video-language pre-training has shown great success especially for retrieval. Yet, existing video-language transformer models do not explicitly fine-grained semantic align. In this work, we present Object-aware Transformers, an object-centric approach that extends video-language transformer to incorporate object representations. The key idea is to leverage the bounding boxes and object tags to guide the training process. We evaluate our model on three standard sub-tasks of video-text matching on four widely used benchmarks. We also provide deep analysis and detailed ablation about the proposed method. We show clear improvement in performance across all tasks and datasets considered, demonstrating the value of a model that incorporates object representations into a video-language architecture. The code will be released at \url{https://github.com/FingerRec/OA-Transformer}.
In this paper, we propose a single UniFied transfOrmer (UFO), which is capable of processing either unimodal inputs (e.g., image or language) or multimodal inputs (e.g., the concatenation of the image and the question), for vision-language (VL) representation learning. Existing approaches typically design an individual network for each modality and/or a specific fusion network for multimodal tasks. To simplify the network architecture, we use a single transformer network and enforce multi-task learning during VL pre-training, which includes the image-text contrastive loss, image-text matching loss, and masked language modeling loss based on the bidirectional and the seq2seq attention mask. The same transformer network is used as the image encoder, the text encoder, or the fusion network in different pre-training tasks. Empirically, we observe less conflict among different tasks and achieve new state of the arts on visual question answering, COCO image captioning (cross-entropy optimization) and nocaps (in SPICE). On other downstream tasks, e.g., image-text retrieval, we also achieve competitive performance.
We present a method for matching a text sentence from a given corpus to a given video clip and vice versa. Traditionally video and text matching is done by learning a shared embedding space and the encoding of one modality is independent of the other. In this work, we encode the dataset data in a way that takes into account the query's relevant information. The power of the method is demonstrated to arise from pooling the interaction data between words and frames. Since the encoding of the video clip depends on the sentence compared to it, the representation needs to be recomputed for each potential match. To this end, we propose an efficient shallow neural network. Its training employs a hierarchical triplet loss that is extendable to paragraph/video matching. The method is simple, provides explainability, and achieves state-of-the-art results for both sentence-clip and video-text by a sizable margin across five different datasets: ActivityNet, DiDeMo, YouCook2, MSR-VTT, and LSMDC. We also show that our conditioned representation can be transferred to video-guided machine translation, where we improved the current results on VATEX. Source code is available at https://github.com/AmeenAli/VideoMatch.
Unsupervised pre-training is now the predominant approach for both text and speech understanding. Self-attention models pre-trained on large amounts of unannotated data have been hugely successful when fine-tuned on downstream tasks from a variety of domains and languages. This paper takes the universality of unsupervised language pre-training one step further, by unifying speech and text pre-training within a single model. We build a single encoder with the BERT objective on unlabeled text together with the w2v-BERT objective on unlabeled speech. To further align our model representations across modalities, we leverage alignment losses, specifically Translation Language Modeling (TLM) and Speech Text Matching (STM) that make use of supervised speech-text recognition data. We demonstrate that incorporating both speech and text data during pre-training can significantly improve downstream quality on CoVoST~2 speech translation, by around 1 BLEU compared to single-modality pre-trained models, while retaining close to SotA performance on LibriSpeech and SpeechStew ASR tasks. On four GLUE tasks and text-normalization, we observe evidence of capacity limitations and interference between the two modalities, leading to degraded performance compared to an equivalent text-only model, while still being competitive with BERT. Through extensive empirical analysis we also demonstrate the importance of the choice of objective function for speech pre-training, and the beneficial effect of adding additional supervised signals on the quality of the learned representations.
Video captioning is a challenging task since it requires generating sentences describing various diverse and complex videos. Existing video captioning models lack adequate visual representation due to the neglect of the existence of gaps between videos and texts. To bridge this gap, in this paper, we propose a CLIP4Caption framework that improves video captioning based on a CLIP-enhanced video-text matching network (VTM). This framework is taking full advantage of the information from both vision and language and enforcing the model to learn strongly text-correlated video features for text generation. Besides, unlike most existing models using LSTM or GRU as the sentence decoder, we adopt a Transformer structured decoder network to effectively learn the long-range visual and language dependency. Additionally, we introduce a novel ensemble strategy for captioning tasks. Experimental results demonstrate the effectiveness of our method on two datasets: 1) on MSR-VTT dataset, our method achieved a new state-of-the-art result with a significant gain of up to 10% in CIDEr; 2) on the private test data, our method ranking 2nd place in the ACM MM multimedia grand challenge 2021: Pre-training for Video Understanding Challenge. It is noted that our model is only trained on the MSR-VTT dataset.
Although conceptualization has been widely studied in semantics and knowledge representation, it is still challenging to find the most accurate concept phrases to characterize the main idea of a text snippet on the fast-growing social media. This is partly attributed to the fact that most knowledge bases contain general terms of the world, such as trees and cars, which do not have the defining power or are not interesting enough to social media app users. Another reason is that the intricacy of natural language allows the use of tense, negation and grammar to change the logic or emphasis of language, thus conveying completely different meanings. In this paper, we present TAG, a high-quality concept matching dataset consisting of 10,000 labeled pairs of fine-grained concepts and web-styled natural language sentences, mined from the open-domain social media. The concepts we consider represent the trending interests of online users. Associated with TAG is a concept graph of these fine-grained concepts and entities to provide the structural context information. We evaluate a wide range of popular neural text matching models as well as pre-trained language models on TAG, and point out their insufficiency to tag social media content with the most appropriate concept. We further propose a novel graph-graph matching method that demonstrates superior abstraction and generalization performance by better utilizing both the structural context in the concept graph and logic interactions between semantic units in the sentence via syntactic dependency parsing. We open-source both the TAG dataset and the proposed methods to facilitate further research.
The task of image-text matching aims to map representations from different modalities into a common joint visual-textual embedding. However, the most widely used datasets for this task, MSCOCO and Flickr30K, are actually image captioning datasets that offer a very limited set of relationships between images and sentences in their ground-truth annotations. This limited ground truth information forces us to use evaluation metrics based on binary relevance: given a sentence query we consider only one image as relevant. However, many other relevant images or captions may be present in the dataset. In this work, we propose two metrics that evaluate the degree of semantic relevance of retrieved items, independently of their annotated binary relevance. Additionally, we incorporate a novel strategy that uses an image captioning metric, CIDEr, to define a Semantic Adaptive Margin (SAM) to be optimized in a standard triplet loss. By incorporating our formulation to existing models, a \emph{large} improvement is obtained in scenarios where available training data is limited. We also demonstrate that the performance on the annotated image-caption pairs is maintained while improving on other non-annotated relevant items when employing the full training set. Code with our metrics and adaptive margin formulation will be made public.
Semantic annotation of long texts, such as novels, remains an open challenge in Natural Language Processing (NLP). This research investigates the problem of detecting person entities and assigning them unique identities, i.e., recognizing people (especially main characters) in novels. We prepared a method for person entity linkage (named entity recognition and disambiguation) and new testing datasets. The datasets comprise 1,300 sentences from 13 classic novels of different genres that a novel reader had manually annotated. Our process of identifying literary characters in a text, implemented in protagonistTagger, comprises two stages: (1) named entity recognition (NER) of persons, (2) named entity disambiguation (NED) - matching each recognized person with the literary character's full name, based on approximate text matching. The protagonistTagger achieves both precision and recall of above 83% on the prepared testing sets. Finally, we gathered a corpus of 13 full-text novels tagged with protagonistTagger that comprises more than 35,000 mentions of literary characters.
Knowledge enhanced pre-trained language models (K-PLMs) are shown to be effective for many public tasks in the literature but few of them have been successfully applied in practice. To address this problem, we propose K-AID, a systematic approach that includes a low-cost knowledge acquisition process for acquiring domain knowledge, an effective knowledge infusion module for improving model performance, and a knowledge distillation component for reducing the model size and deploying K-PLMs on resource-restricted devices (e.g., CPU) for real-world application. Importantly, instead of capturing entity knowledge like the majority of existing K-PLMs, our approach captures relational knowledge, which contributes to better-improving sentence-level text classification and text matching tasks that play a key role in question answering (QA). We conducted a set of experiments on five text classification tasks and three text matching tasks from three domains, namely E-commerce, Government, and Film&TV, and performed online A/B tests in E-commerce. Experimental results show that our approach is able to achieve substantial improvement on sentence-level question answering tasks and bring beneficial business value in industrial settings.
The canonical approach to video action recognition dictates a neural model to do a classic and standard 1-of-N majority vote task. They are trained to predict a fixed set of predefined categories, limiting their transferable ability on new datasets with unseen concepts. In this paper, we provide a new perspective on action recognition by attaching importance to the semantic information of label texts rather than simply mapping them into numbers. Specifically, we model this task as a video-text matching problem within a multimodal learning framework, which strengthens the video representation with more semantic language supervision and enables our model to do zero-shot action recognition without any further labeled data or parameters requirements. Moreover, to handle the deficiency of label texts and make use of tremendous web data, we propose a new paradigm based on this multimodal learning framework for action recognition, which we dub "pre-train, prompt and fine-tune". This paradigm first learns powerful representations from pre-training on a large amount of web image-text or video-text data. Then it makes the action recognition task to act more like pre-training problems via prompt engineering. Finally, it end-to-end fine-tunes on target datasets to obtain strong performance. We give an instantiation of the new paradigm, ActionCLIP, which not only has superior and flexible zero-shot/few-shot transfer ability but also reaches a top performance on general action recognition task, achieving 83.8% top-1 accuracy on Kinetics-400 with a ViT-B/16 as the backbone. Code is available at https://github.com/sallymmx/ActionCLIP.git
Both image-caption pairs and translation pairs provide the means to learn deep representations of and connections between languages. We use both types of pairs in MURAL (MUltimodal, MUltitask Representations Across Languages), a dual encoder that solves two tasks: 1) image-text matching and 2) translation pair matching. By incorporating billions of translation pairs, MURAL extends ALIGN (Jia et al. PMLR'21)--a state-of-the-art dual encoder learned from 1.8 billion noisy image-text pairs. When using the same encoders, MURAL's performance matches or exceeds ALIGN's cross-modal retrieval performance on well-resourced languages across several datasets. More importantly, it considerably improves performance on under-resourced languages, showing that text-text learning can overcome a paucity of image-caption examples for these languages. On the Wikipedia Image-Text dataset, for example, MURAL-base improves zero-shot mean recall by 8.1% on average for eight under-resourced languages and by 6.8% on average when fine-tuning. We additionally show that MURAL's text representations cluster not only with respect to genealogical connections but also based on areal linguistics, such as the Balkan Sprachbund.
Recent advancements in deep learning techniques have transformed the area of semantic text matching. However, most state-of-the-art models are designed to operate with short documents such as tweets, user reviews, comments, etc. These models have fundamental limitations when applied to long-form documents such as scientific papers, legal documents, and patents. When handling such long documents, there are three primary challenges: (i) the presence of different contexts for the same word throughout the document, (ii) small sections of contextually similar text between two documents, but dissimilar text in the remaining parts (this defies the basic understanding of "similarity"), and (iii) the coarse nature of a single global similarity measure which fails to capture the heterogeneity of the document content. In this paper, we describe CoLDE: Contrastive Long Document Encoder - a transformer-based framework that addresses these challenges and allows for interpretable comparisons of long documents. CoLDE uses unique positional embeddings and a multi-headed chunkwise attention layer in conjunction with a supervised contrastive learning framework to capture similarity at three different levels: (i) high-level similarity scores between a pair of documents, (ii) similarity scores between different sections within and across documents, and (iii) similarity scores between different chunks in the same document and across other documents. These fine-grained similarity scores aid in better interpretability. We evaluate CoLDE on three long document datasets namely, ACL Anthology publications, Wikipedia articles, and USPTO patents. Besides outperforming the state-of-the-art methods on the document matching task, CoLDE is also robust to changes in document length and text perturbations and provides interpretable results.
Semantic text matching is a critical problem in information retrieval. Recently, deep learning techniques have been widely used in this area and obtained significant performance improvements. However, most models are black boxes and it is hard to understand what happened in the matching process, due to the poor interpretability of deep learning. This paper aims at tackling this problem. The key idea is to test whether existing deep text matching methods satisfy some fundamental heuristics in information retrieval. Specifically, four heuristics are used in our study, i.e., term frequency constraint, term discrimination constraint, length normalization constraints, and TF-length constraint. Since deep matching models usually contain many parameters, it is difficult to conduct a theoretical study for these complicated functions. In this paper, We propose an empirical testing method. Specifically, We first construct some queries and documents to make them satisfy the assumption in a constraint, and then test to which extend a deep text matching model trained on the original dataset satisfies the corresponding constraint. Besides, a famous attribution based interpretation method, namely integrated gradient, is adopted to conduct detailed analysis and guide for feasible improvement. Experimental results on LETOR 4.0 and MS Marco show that all the investigated deep text matching methods, both representation and interaction based methods, satisfy the above constraints with high probabilities in statistics. We further extend these constraints to the semantic settings, which are shown to be better satisfied for all the deep text matching models. These empirical findings give clear understandings on why deep text matching models usually perform well in information retrieval. We believe the proposed evaluation methodology will be useful for testing future deep text matching models.
Video-text retrieval is an important yet challenging task in vision-language understanding, which aims to learn a joint embedding space where related video and text instances are close to each other. Most current works simply measure the video-text similarity based on video-level and text-level embeddings. However, the neglect of more fine-grained or local information causes the problem of insufficient representation. Some works exploit the local details by disentangling sentences, but overlook the corresponding videos, causing the asymmetry of video-text representation. To address the above limitations, we propose a Hierarchical Alignment Network (HANet) to align different level representations for video-text matching. Specifically, we first decompose video and text into three semantic levels, namely event (video and text), action (motion and verb), and entity (appearance and noun). Based on these, we naturally construct hierarchical representations in the individual-local-global manner, where the individual level focuses on the alignment between frame and word, local level focuses on the alignment between video clip and textual context, and global level focuses on the alignment between the whole video and text. Different level alignments capture fine-to-coarse correlations between video and text, as well as take the advantage of the complementary information among three semantic levels. Besides, our HANet is also richly interpretable by explicitly learning key semantic concepts. Extensive experiments on two public datasets, namely MSR-VTT and VATEX, show the proposed HANet outperforms other state-of-the-art methods, which demonstrates the effectiveness of hierarchical representation and alignment. Our code is publicly available.
Shape-Text matching is an important task of high-level shape understanding. Current methods mainly represent a 3D shape as multiple 2D rendered views, which obviously can not be understood well due to the structural ambiguity caused by self-occlusion in the limited number of views. To resolve this issue, we directly represent 3D shapes as point clouds, and propose to learn joint embedding of point clouds and texts by bidirectional matching between parts from shapes and words from texts. Specifically, we first segment the point clouds into parts, and then leverage optimal transport method to match parts and words in an optimized feature space, where each part is represented by aggregating features of all points within it and each word is abstracted by its contextual information. We optimize the feature space in order to enlarge the similarities between the paired training samples, while simultaneously maximizing the margin between the unpaired ones. Experiments demonstrate that our method achieves a significant improvement in accuracy over the SOTAs on multi-modal retrieval tasks under the Text2Shape dataset. Codes are available at https://github.com/JLUtangchuan/Parts2Words.
Existing neural ranking models follow the text matching paradigm, where document-to-query relevance is estimated through predicting the matching score. Drawing from the rich literature of classical generative retrieval models, we introduce and formalize the paradigm of deep generative retrieval models defined via the cumulative probabilities of generating query terms. This paradigm offers a grounded probabilistic view on relevance estimation while still enabling the use of modern neural architectures. In contrast to the matching paradigm, the probabilistic nature of generative rankers readily offers a fine-grained measure of uncertainty. We adopt several current neural generative models in our framework and introduce a novel generative ranker (T-PGN), which combines the encoding capacity of Transformers with the Pointer Generator Network model. We conduct an extensive set of evaluation experiments on passage retrieval, leveraging the MS MARCO Passage Re-ranking and TREC Deep Learning 2019 Passage Re-ranking collections. Our results show the significantly higher performance of the T-PGN model when compared with other generative models. Lastly, we demonstrate that exploiting the uncertainty information of deep generative rankers opens new perspectives to query/collection understanding, and significantly improves the cut-off prediction task.
Image-text matching plays a central role in bridging the semantic gap between vision and language. The key point to achieve precise visual-semantic alignment lies in capturing the fine-grained cross-modal correspondence between image and text. Most previous methods rely on single-step reasoning to discover the visual-semantic interactions, which lacks the ability of exploiting the multi-level information to locate the hierarchical fine-grained relevance. Different from them, in this work, we propose a step-wise hierarchical alignment network (SHAN) that decomposes image-text matching into multi-step cross-modal reasoning process. Specifically, we first achieve local-to-local alignment at fragment level, following by performing global-to-local and global-to-global alignment at context level sequentially. This progressive alignment strategy supplies our model with more complementary and sufficient semantic clues to understand the hierarchical correlations between image and text. The experimental results on two benchmark datasets demonstrate the superiority of our proposed method.
Conventional approaches to image-text retrieval mainly focus on indexing visual objects appearing in pictures but ignore the interactions between these objects. Such objects occurrences and interactions are equivalently useful and important in this field as they are usually mentioned in the text. Scene graph presentation is a suitable method for the image-text matching challenge and obtained good results due to its ability to capture the inter-relationship information. Both images and text are represented in scene graph levels and formulate the retrieval challenge as a scene graph matching challenge. In this paper, we introduce the Local and Global Scene Graph Matching (LGSGM) model that enhances the state-of-the-art method by integrating an extra graph convolution network to capture the general information of a graph. Specifically, for a pair of scene graphs of an image and its caption, two separate models are used to learn the features of each graph's nodes and edges. Then a Siamese-structure graph convolution model is employed to embed graphs into vector forms. We finally combine the graph-level and the vector-level to calculate the similarity of this image-text pair. The empirical experiments show that our enhancement with the combination of levels can improve the performance of the baseline method by increasing the recall by more than 10% on the Flickr30k dataset.