We propose a cross-modal attention distillation framework to train a dual-encoder model for vision-language understanding tasks, such as visual reasoning and visual question answering. Dual-encoder models have a faster inference speed than fusion-encoder models and enable the pre-computation of images and text during inference. However, the shallow interaction module used in dual-encoder models is insufficient to handle complex vision-language understanding tasks. In order to learn deep interactions of images and text, we introduce cross-modal attention distillation, which uses the image-to-text and text-to-image attention distributions of a fusion-encoder model to guide the training of our dual-encoder model. In addition, we show that applying the cross-modal attention distillation for both pre-training and fine-tuning stages achieves further improvements. Experimental results demonstrate that the distilled dual-encoder model achieves competitive performance for visual reasoning, visual entailment and visual question answering tasks while enjoying a much faster inference speed than fusion-encoder models. Our code and models will be publicly available at https://github.com/kugwzk/Distilled-DualEncoder.
Most existing vision-language pre-training methods focus on understanding tasks and use BERT-like objectives (masked language modeling and image-text matching) during pretraining. Although they perform well in many understanding downstream tasks, e.g., visual question answering, image-text retrieval and visual entailment, they do not possess the ability to generate. To tackle this problem, we propose Unified multimodal pre-training for both Vision-Language understanding and generation (UniVL). The proposed UniVL is capable of handling both understanding tasks and generative tasks. We augment existing pretraining paradigms that only use random masks with causal masks, i.e., triangular masks that mask out future tokens, such that the pre-trained models can have autoregressive generation abilities by design. We formulate several previous understanding tasks as a text generation task and propose to use prompt-based method for fine-tuning on different downstream tasks. Our experiments show that there is a trade-off between understanding tasks and generation tasks while using the same model, and a feasible way to improve both tasks is to use more data. Our UniVL framework attains comparable performance to recent vision-language pre-training methods on both understanding tasks and generation tasks. Moreover, we demostrate that prompt-based finetuning is more data-efficient - it outperforms discriminative methods in few-shot scenarios.
Most existing Vision-and-Language (V&L) models rely on pre-trained visual encoders, using a relatively small set of manually-annotated data (as compared to web-crawled data), to perceive the visual world. However, it has been observed that large-scale pretraining usually can result in better generalization performance, e.g., CLIP (Contrastive Language-Image Pre-training), trained on a massive amount of image-caption pairs, has shown a strong zero-shot capability on various vision tasks. To further study the advantage brought by CLIP, we propose to use CLIP as the visual encoder in various V&L models in two typical scenarios: 1) plugging CLIP into task-specific fine-tuning; 2) combining CLIP with V&L pre-training and transferring to downstream tasks. We show that CLIP significantly outperforms widely-used visual encoders trained with in-domain annotated data, such as BottomUp-TopDown. We achieve competitive or better results on diverse V&L tasks, while establishing new state-of-the-art results on Visual Question Answering, Visual Entailment, and V&L Navigation tasks. We release our code at https://github.com/clip-vil/CLIP-ViL.
Vision-Language Pre-training (VLP) aims to learn multi-modal representations from image-text pairs and serves for downstream vision-language tasks in a fine-tuning fashion. The dominant VLP models adopt a CNN-Transformer architecture, which embeds images with a CNN, and then aligns images and text with a Transformer. Visual relationship between visual contents plays an important role in image understanding and is the basic for inter-modal alignment learning. However, CNNs have limitations in visual relation learning due to local receptive field's weakness in modeling long-range dependencies. Thus the two objectives of learning visual relation and inter-modal alignment are encapsulated in the same Transformer network. Such design might restrict the inter-modal alignment learning in the Transformer by ignoring the specialized characteristic of each objective. To tackle this, we propose a fully Transformer visual embedding for VLP to better learn visual relation and further promote inter-modal alignment. Specifically, we propose a metric named Inter-Modality Flow (IMF) to measure the interaction between vision and language modalities (i.e., inter-modality). We also design a novel masking optimization mechanism named Masked Feature Regression (MFR) in Transformer to further promote the inter-modality learning. To the best of our knowledge, this is the first study to explore the benefit of Transformer for visual feature learning in VLP. We verify our method on a wide range of vision-language tasks, including Image-Text Retrieval, Visual Question Answering (VQA), Visual Entailment and Visual Reasoning. Our approach not only outperforms the state-of-the-art VLP performance, but also shows benefits on the IMF metric.
While sophisticated Visual Question Answering models have achieved remarkable success, they tend to answer questions only according to superficial correlations between question and answer. Several recent approaches have been developed to address this language priors problem. However, most of them predict the correct answer according to one best output without checking the authenticity of answers. Besides, they only explore the interaction between image and question, ignoring the semantics of candidate answers. In this paper, we propose a select-and-rerank (SAR) progressive framework based on Visual Entailment. Specifically, we first select the candidate answers relevant to the question or the image, then we rerank the candidate answers by a visual entailment task, which verifies whether the image semantically entails the synthetic statement of the question and each candidate answer. Experimental results show the effectiveness of our proposed framework, which establishes a new state-of-the-art accuracy on VQA-CP v2 with a 7.55% improvement.
Large-scale pre-training has recently revolutionized vision-and-language (VL) research. Models such as LXMERT and UNITER have significantly lifted the state of the art over a wide range of VL tasks. However, the large number of parameters in such models hinders their application in practice. In parallel, work on the lottery ticket hypothesis (LTH) has shown that deep neural networks contain small matching subnetworks that can achieve on par or even better performance than the dense networks when trained in isolation. In this work, we perform the first empirical study to assess whether such trainable subnetworks also exist in pre-trained VL models. We use UNITER as the main testbed (also test on LXMERT and ViLT), and consolidate 7 representative VL tasks for experiments, including visual question answering, visual commonsense reasoning, visual entailment, referring expression comprehension, image-text retrieval, GQA, and NLVR$^2$. Through comprehensive analysis, we summarize our main findings as follows. ($i$) It is difficult to find subnetworks that strictly match the performance of the full model. However, we can find "relaxed" winning tickets at 50%-70% sparsity that maintain 99% of the full accuracy. ($ii$) Subnetworks found by task-specific pruning transfer reasonably well to the other tasks, while those found on the pre-training tasks at 60%/70% sparsity transfer universally, matching 98%/96% of the full accuracy on average over all the tasks. ($iii$) Besides UNITER, other models such as LXMERT and ViLT can also play lottery tickets. However, the highest sparsity we can achieve for ViLT is far lower than LXMERT and UNITER (30% vs. 70%). ($iv$) LTH also remains relevant when using other training methods (e.g., adversarial training).
We study joint learning of Convolutional Neural Network (CNN) and Transformer for vision-language pre-training (VLPT) which aims to learn cross-modal alignments from millions of image-text pairs. State-of-the-art approaches extract salient image regions and align regions with words step-by-step. As region-based visual features usually represent parts of an image, it is challenging for existing vision-language models to fully understand the semantics from paired natural languages. In this paper, we propose SOHO to "See Out of tHe bOx" that takes a whole image as input, and learns vision-language representation in an end-to-end manner. SOHO does not require bounding box annotations which enables inference 10 times faster than region-based approaches. In particular, SOHO learns to extract comprehensive yet compact image features through a visual dictionary (VD) that facilitates cross-modal understanding. VD is designed to represent consistent visual abstractions of similar semantics. It is updated on-the-fly and utilized in our proposed pre-training task Masked Visual Modeling (MVM). We conduct experiments on four well-established vision-language tasks by following standard VLPT settings. In particular, SOHO achieves absolute gains of 2.0% R@1 score on MSCOCO text retrieval 5k test split, 1.5% accuracy on NLVR$^2$ test-P split, 6.7% accuracy on SNLI-VE test split, respectively.
We present VILLA, the first known effort on large-scale adversarial training for vision-and-language (V+L) representation learning. VILLA consists of two training stages: (i) task-agnostic adversarial pre-training; followed by (ii) task-specific adversarial finetuning. Instead of adding adversarial perturbations on image pixels and textual tokens, we propose to perform adversarial training in the embedding space of each modality. To enable large-scale training, we adopt the "free" adversarial training strategy, and combine it with KL-divergence-based regularization to promote higher invariance in the embedding space. We apply VILLA to current best-performing V+L models, and achieve new state of the art on a wide range of tasks, including Visual Question Answering, Visual Commonsense Reasoning, Image-Text Retrieval, Referring Expression Comprehension, Visual Entailment, and NLVR2.
Joint image-text embedding is the bedrock for most Vision-and-Language (V+L) tasks, where multimodality inputs are simultaneously processed for joint visual and textual understanding. In this paper, we introduce UNITER, a UNiversal Image-TExt Representation, learned through large-scale pre-training over four image-text datasets (COCO, Visual Genome, Conceptual Captions, and SBU Captions), which can power heterogeneous downstream V+L tasks with joint multimodal embeddings. We design four pre-training tasks: Masked Language Modeling (MLM), Masked Region Modeling (MRM, with three variants), Image-Text Matching (ITM), and Word-Region Alignment (WRA). Different from previous work that applies joint random masking to both modalities, we use conditional masking on pre-training tasks (i.e., masked language/region modeling is conditioned on full observation of image/text). In addition to ITM for global image-text alignment, we also propose WRA via the use of Optimal Transport (OT) to explicitly encourage fine-grained alignment between words and image regions during pre-training. Comprehensive analysis shows that both conditional masking and OT-based WRA contribute to better pre-training. We also conduct a thorough ablation study to find an optimal combination of pre-training tasks. Extensive experiments show that UNITER achieves new state of the art across six V+L tasks (over nine datasets), including Visual Question Answering, Image-Text Retrieval, Referring Expression Comprehension, Visual Commonsense Reasoning, Visual Entailment, and NLVR$^2$. Code is available at https://github.com/ChenRocks/UNITER.
Existing visual reasoning datasets such as Visual Question Answering (VQA), often suffer from biases conditioned on the question, image or answer distributions. The recently proposed CLEVR dataset addresses these limitations and requires fine-grained reasoning but the dataset is synthetic and consists of similar objects and sentence structures across the dataset. In this paper, we introduce a new inference task, Visual Entailment (VE) - consisting of image-sentence pairs whereby a premise is defined by an image, rather than a natural language sentence as in traditional Textual Entailment tasks. The goal of a trained VE model is to predict whether the image semantically entails the text. To realize this task, we build a dataset SNLI-VE based on the Stanford Natural Language Inference corpus and Flickr30k dataset. We evaluate various existing VQA baselines and build a model called Explainable Visual Entailment (EVE) system to address the VE task. EVE achieves up to 71% accuracy and outperforms several other state-of-the-art VQA based models. Finally, we demonstrate the explainability of EVE through cross-modal attention visualizations. The SNLI-VE dataset is publicly available at https://github.com/ necla-ml/SNLI-VE.
We introduce a new inference task - Visual Entailment (VE) - which differs from traditional Textual Entailment (TE) tasks whereby a premise is defined by an image, rather than a natural language sentence as in TE tasks. A novel dataset SNLI-VE (publicly available at https://github.com/necla-ml/SNLI-VE) is proposed for VE tasks based on the Stanford Natural Language Inference corpus and Flickr30k. We introduce a differentiable architecture called the Explainable Visual Entailment model (EVE) to tackle the VE problem. EVE and several other state-of-the-art visual question answering (VQA) based models are evaluated on the SNLI-VE dataset, facilitating grounded language understanding and providing insights on how modern VQA based models perform.
We introduce Segment-Phrase Table (SPT), a large collection of bijective associations between textual phrases and their corresponding segmentations. Leveraging recent progress in object recognition and natural language semantics, we show how we can successfully build a high-quality segment-phrase table using minimal human supervision. More importantly, we demonstrate the unique value unleashed by this rich bimodal resource, for both vision as well as natural language understanding. First, we show that fine-grained textual labels facilitate contextual reasoning that helps in satisfying semantic constraints across image segments. This feature enables us to achieve state-of-the-art segmentation results on benchmark datasets. Next, we show that the association of high-quality segmentations to textual phrases aids in richer semantic understanding and reasoning of these textual phrases. Leveraging this feature, we motivate the problem of visual entailment and visual paraphrasing, and demonstrate its utility on a large dataset.