Visual and linguistic pre-training aims to learn vision and language representations together, which can be transferred to visual-linguistic downstream tasks. However, there exists semantic confusion between language and vision during the pre-training stage. Moreover, current pre-trained models tend to take lots of computation resources for fine-tuning when transferred to downstream tasks. In this work, we present a simple but effective approach for learning Contrastive and Adaptive representations of Vision and Language, namely CAVL. Specifically, we introduce a pair-wise contrastive loss to learn alignments between the whole sentence and each image in the same batch during the pre-training process. At the fine-tuning stage, we introduce two lightweight adaptation networks to reduce model parameters and increase training speed for saving computation resources. We evaluate our CAVL on six main downstream tasks, including Visual Question Answering (VQA), Visual Commonsense Reasoning (VCR), Natural Language for Visual Reasoning (NLVR), Region-to-Phrase Grounding (RPG), Text-to-Image Retrieval (TIR), and Zero-shot Text-to-Image Retrieval (ZS-TIR). Compared to baselines, we achieve superior performance and reduce the fine-tuning time by a large margin (in particular, 76.17%). Extensive experiments and ablation studies demonstrate the efficiency of contrastive pre-training and adaptive fine-tuning proposed in our CAVL.
Deep neural networks have been critical in the task of Visual Question Answering (VQA), with research traditionally focused on improving model accuracy. Recently, however, there has been a trend towards evaluating the robustness of these models against adversarial attacks. This involves assessing the accuracy of VQA models under increasing levels of noise in the input, which can target either the image or the proposed query question, dubbed the main question. However, there is currently a lack of proper analysis of this aspect of VQA. This work proposes a new method that utilizes semantically related questions, referred to as basic questions, acting as noise to evaluate the robustness of VQA models. It is hypothesized that as the similarity of a basic question to the main question decreases, the level of noise increases. To generate a reasonable noise level for a given main question, a pool of basic questions is ranked based on their similarity to the main question, and this ranking problem is cast as a LASSO optimization problem. Additionally, this work proposes a novel robustness measure, R_score, and two basic question datasets to standardize the analysis of VQA model robustness. The experimental results demonstrate that the proposed evaluation method effectively analyzes the robustness of VQA models. Moreover, the experiments show that in-context learning with a chain of basic questions can enhance model accuracy.
Adapters present a promising solution to the catastrophic forgetting problem in continual learning. However, training independent Adapter modules for every new task misses an opportunity for cross-task knowledge transfer. We propose Improvise to Initialize (I2I), a continual learning algorithm that initializes Adapters for incoming tasks by distilling knowledge from previously-learned tasks' Adapters. We evaluate I2I on CLiMB, a multimodal continual learning benchmark, by conducting experiments on sequences of visual question answering tasks. Adapters trained with I2I consistently achieve better task accuracy than independently-trained Adapters, demonstrating that our algorithm facilitates knowledge transfer between task Adapters. I2I also results in better cross-task knowledge transfer than the state-of-the-art AdapterFusion without incurring the associated parametric cost.
Visual question answering (VQA) is a critical multimodal task in which an agent must answer questions according to the visual cue. Unfortunately, language bias is a common problem in VQA, which refers to the model generating answers only by associating with the questions while ignoring the visual content, resulting in biased results. We tackle the language bias problem by proposing a self-supervised counterfactual metric learning (SC-ML) method to focus the image features better. SC-ML can adaptively select the question-relevant visual features to answer the question, reducing the negative influence of question-irrelevant visual features on inferring answers. In addition, question-irrelevant visual features can be seamlessly incorporated into counterfactual training schemes to further boost robustness. Extensive experiments have proved the effectiveness of our method with improved results on the VQA-CP dataset. Our code will be made publicly available.
Medical Visual Question Answering (VQA) systems play a supporting role to understand clinic-relevant information carried by medical images. The questions to a medical image include two categories: close-end (such as Yes/No question) and open-end. To obtain answers, the majority of the existing medical VQA methods relies on classification approaches, while a few works attempt to use generation approaches or a mixture of the two. The classification approaches are relatively simple but perform poorly on long open-end questions. To bridge this gap, in this paper, we propose a new Transformer based framework for medical VQA (named as Q2ATransformer), which integrates the advantages of both the classification and the generation approaches and provides a unified treatment for the close-end and open-end questions. Specifically, we introduce an additional Transformer decoder with a set of learnable candidate answer embeddings to query the existence of each answer class to a given image-question pair. Through the Transformer attention, the candidate answer embeddings interact with the fused features of the image-question pair to make the decision. In this way, despite being a classification-based approach, our method provides a mechanism to interact with the answer information for prediction like the generation-based approaches. On the other hand, by classification, we mitigate the task difficulty by reducing the search space of answers. Our method achieves new state-of-the-art performance on two medical VQA benchmarks. Especially, for the open-end questions, we achieve 79.19% on VQA-RAD and 54.85% on PathVQA, with 16.09% and 41.45% absolute improvements, respectively.
In this paper, we propose a novel multi-modal framework for Scene Text Visual Question Answering (STVQA), which requires models to read scene text in images for question answering. Apart from text or visual objects, which could exist independently, scene text naturally links text and visual modalities together by conveying linguistic semantics while being a visual object in an image simultaneously. Different to conventional STVQA models which take the linguistic semantics and visual semantics in scene text as two separate features, in this paper, we propose a paradigm of "Locate Then Generate" (LTG), which explicitly unifies this two semantics with the spatial bounding box as a bridge connecting them. Specifically, at first, LTG locates the region in an image that may contain the answer words with an answer location module (ALM) consisting of a region proposal network and a language refinement network, both of which can transform to each other with one-to-one mapping via the scene text bounding box. Next, given the answer words selected by ALM, LTG generates a readable answer sequence with an answer generation module (AGM) based on a pre-trained language model. As a benefit of the explicit alignment of the visual and linguistic semantics, even without any scene text based pre-training tasks, LTG can boost the absolute accuracy by +6.06% and +6.92% on the TextVQA dataset and the ST-VQA dataset respectively, compared with a non-pre-training baseline. We further demonstrate that LTG effectively unifies visual and text modalities through the spatial bounding box connection, which is underappreciated in previous methods.
Trojan attacks embed perturbations in input data leading to malicious behavior in neural network models. A combination of various Trojans in different modalities enables an adversary to mount a sophisticated attack on multimodal learning such as Visual Question Answering (VQA). However, multimodal Trojans in conventional methods are susceptible to parameter adjustment during processes such as fine-tuning. To this end, we propose an instance-level multimodal Trojan attack on VQA that efficiently adapts to fine-tuned models through a dual-modality adversarial learning method. This method compromises two specific neurons in a specific perturbation layer in the pretrained model to produce overly large neuron activations. Then, a malicious correlation between these overactive neurons and the malicious output of a fine-tuned model is established through adversarial learning. Extensive experiments are conducted using the VQA-v2 dataset, based on a wide range of metrics including sample efficiency, stealthiness, and robustness. The proposed attack demonstrates enhanced performance with diverse vision and text Trojans tailored for each sample. We demonstrate that the proposed attack can be efficiently adapted to different fine-tuned models, by injecting only a few shots of Trojan samples. Moreover, we investigate the attack performance under conventional defenses, where the defenses cannot effectively mitigate the attack.
There has been a recent explosion of computer vision models which perform many tasks and are composed of an image encoder (usually a ViT) and an autoregressive decoder (usually a Transformer). However, most of this work simply presents one system and its results, leaving many questions regarding design decisions and trade-offs of such systems unanswered. In this work, we aim to provide such answers. We take a close look at autoregressive decoders for multi-task learning in multimodal computer vision, including classification, captioning, visual question answering, and optical character recognition. Through extensive systematic experiments, we study the effects of task and data mixture, training and regularization hyperparameters, conditioning type and specificity, modality combination, and more. Importantly, we compare these to well-tuned single-task baselines to highlight the cost incurred by multi-tasking. A key finding is that a small decoder learned on top of a frozen pretrained encoder works surprisingly well. We call this setup locked-image tuning with decoder (LiT-decoder). It can be seen as teaching a decoder to interact with a pretrained vision model via natural language.
The development of language models have moved from encoder-decoder to decoder-only designs. In addition, we observe that the two most popular multimodal tasks, the generative and contrastive tasks, are nontrivial to accommodate in one architecture, and further need adaptations for downstream tasks. We propose a novel paradigm of training with a decoder-only model for multimodal tasks, which is surprisingly effective in jointly learning of these disparate vision-language tasks. This is done with a simple model, called MaMMUT. It consists of a single vision encoder and a text decoder, and is able to accommodate contrastive and generative learning by a novel two-pass approach on the text decoder. We demonstrate that joint learning of these diverse objectives is simple, effective, and maximizes the weight-sharing of the model across these tasks. Furthermore, the same architecture enables straightforward extensions to open-vocabulary object detection and video-language tasks. The model tackles a diverse range of tasks, while being modest in capacity. Our model achieves the state of the art on image-text and text-image retrieval, video question answering and open-vocabulary detection tasks, outperforming much larger and more extensively trained foundational models. It shows very competitive results on VQA and Video Captioning, especially considering its capacity. Ablations confirm the flexibility and advantages of our approach.
Visual Question Answering (VQA) is a complex task requiring large datasets and expensive training. Neural Module Networks (NMN) first translate the question to a reasoning path, then follow that path to analyze the image and provide an answer. We propose an NMN method that relies on predefined cross-modal embeddings to ``warm start'' learning on the GQA dataset, then focus on Curriculum Learning (CL) as a way to improve training and make a better use of the data. Several difficulty criteria are employed for defining CL methods. We show that by an appropriate selection of the CL method the cost of training and the amount of training data can be greatly reduced, with a limited impact on the final VQA accuracy. Furthermore, we introduce intermediate losses during training and find that this allows to simplify the CL strategy.
While few-shot learning as a transfer learning paradigm has gained significant traction for scenarios with limited data, it has primarily been explored in the context of building unimodal and unilingual models. Furthermore, a significant part of the existing literature in the domain of few-shot multitask learning perform in-context learning which requires manually generated prompts as the input, yielding varying outcomes depending on the level of manual prompt-engineering. In addition, in-context learning suffers from substantial computational, memory, and storage costs which eventually leads to high inference latency because it involves running all of the prompt's examples through the model every time a prediction is made. In contrast, methods based on the transfer learning via the fine-tuning paradigm avoid the aforementioned issues at a one-time cost of fine-tuning weights on a per-task basis. However, such methods lack exposure to few-shot multimodal multitask learning. In this paper, we propose few-shot learning for a multimodal multitask multilingual (FM3) setting by adapting pre-trained vision and language models using task-specific hypernetworks and contrastively fine-tuning them to enable few-shot learning. FM3's architecture combines the best of both worlds of in-context and fine-tuning based learning and consists of three major components: (i) multimodal contrastive fine-tuning to enable few-shot learning, (ii) hypernetwork task adaptation to perform multitask learning, and (iii) task-specific output heads to cater to a plethora of diverse tasks. FM3 learns the most prominent tasks in the vision and language domains along with their intersections, namely visual entailment (VE), visual question answering (VQA), and natural language understanding (NLU) tasks such as neural entity recognition (NER) and the GLUE benchmark including QNLI, MNLI, QQP, and SST-2.
Visual Question Answering (VQA) is a task that requires computers to give correct answers for the input questions based on the images. This task can be solved by humans with ease but is a challenge for computers. The VLSP2022-EVJVQA shared task carries the Visual Question Answering task in the multilingual domain on a newly released dataset: UIT-EVJVQA, in which the questions and answers are written in three different languages: English, Vietnamese and Japanese. We approached the challenge as a sequence-to-sequence learning task, in which we integrated hints from pre-trained state-of-the-art VQA models and image features with Convolutional Sequence-to-Sequence network to generate the desired answers. Our results obtained up to 0.3442 by F1 score on the public test set, 0.4210 on the private test set, and placed 3rd in the competition.
We propose MM-REACT, a system paradigm that integrates ChatGPT with a pool of vision experts to achieve multimodal reasoning and action. In this paper, we define and explore a comprehensive list of advanced vision tasks that are intriguing to solve, but may exceed the capabilities of existing vision and vision-language models. To achieve such advanced visual intelligence, MM-REACT introduces a textual prompt design that can represent text descriptions, textualized spatial coordinates, and aligned file names for dense visual signals such as images and videos. MM-REACT's prompt design allows language models to accept, associate, and process multimodal information, thereby facilitating the synergetic combination of ChatGPT and various vision experts. Zero-shot experiments demonstrate MM-REACT's effectiveness in addressing the specified capabilities of interests and its wide application in different scenarios that require advanced visual understanding. Furthermore, we discuss and compare MM-REACT's system paradigm with an alternative approach that extends language models for multimodal scenarios through joint finetuning. Code, demo, video, and visualization are available at https://multimodal-react.github.io/
Despite thousands of researchers, engineers, and artists actively working on improving text-to-image generation models, systems often fail to produce images that accurately align with the text inputs. We introduce TIFA (Text-to-Image Faithfulness evaluation with question Answering), an automatic evaluation metric that measures the faithfulness of a generated image to its text input via visual question answering (VQA). Specifically, given a text input, we automatically generate several question-answer pairs using a language model. We calculate image faithfulness by checking whether existing VQA models can answer these questions using the generated image. TIFA is a reference-free metric that allows for fine-grained and interpretable evaluations of generated images. TIFA also has better correlations with human judgments than existing metrics. Based on this approach, we introduce TIFA v1.0, a benchmark consisting of 4K diverse text inputs and 25K questions across 12 categories (object, counting, etc.). We present a comprehensive evaluation of existing text-to-image models using TIFA v1.0 and highlight the limitations and challenges of current models. For instance, we find that current text-to-image models, despite doing well on color and material, still struggle in counting, spatial relations, and composing multiple objects. We hope our benchmark will help carefully measure the research progress in text-to-image synthesis and provide valuable insights for further research.
Scene Graph Generation (SGG) is a visual understanding task, aiming to describe a scene as a graph of entities and their relationships with each other. Existing works rely on location labels in form of bounding boxes or segmentation masks, increasing annotation costs and limiting dataset expansion. Recognizing that many applications do not require location data, we break this dependency and introduce location-free scene graph generation (LF-SGG). This new task aims at predicting instances of entities, as well as their relationships, without the explicit calculation of their spatial localization. To objectively evaluate the task, the predicted and ground truth scene graphs need to be compared. We solve this NP-hard problem through an efficient branching algorithm. Additionally, we design the first LF-SGG method, Pix2SG, using autoregressive sequence modeling. We demonstrate the effectiveness of our method on three scene graph generation datasets as well as two downstream tasks, image retrieval and visual question answering, and show that our approach is competitive to existing methods while not relying on location cues.
The widely used Fact-based Visual Question Answering (FVQA) dataset contains visually-grounded questions that require information retrieval using common sense knowledge graphs to answer. It has been observed that the original dataset is highly imbalanced and concentrated on a small portion of its associated knowledge graph. We introduce FVQA 2.0 which contains adversarial variants of test questions to address this imbalance. We show that systems trained with the original FVQA train sets can be vulnerable to adversarial samples and we demonstrate an augmentation scheme to reduce this vulnerability without human annotations.
Humans are able to accurately reason in 3D by gathering multi-view observations of the surrounding world. Inspired by this insight, we introduce a new large-scale benchmark for 3D multi-view visual question answering (3DMV-VQA). This dataset is collected by an embodied agent actively moving and capturing RGB images in an environment using the Habitat simulator. In total, it consists of approximately 5k scenes, 600k images, paired with 50k questions. We evaluate various state-of-the-art models for visual reasoning on our benchmark and find that they all perform poorly. We suggest that a principled approach for 3D reasoning from multi-view images should be to infer a compact 3D representation of the world from the multi-view images, which is further grounded on open-vocabulary semantic concepts, and then to execute reasoning on these 3D representations. As the first step towards this approach, we propose a novel 3D concept learning and reasoning (3D-CLR) framework that seamlessly combines these components via neural fields, 2D pre-trained vision-language models, and neural reasoning operators. Experimental results suggest that our framework outperforms baseline models by a large margin, but the challenge remains largely unsolved. We further perform an in-depth analysis of the challenges and highlight potential future directions.
Despite considerable recent progress in Visual Question Answering (VQA) models, inconsistent or contradictory answers continue to cast doubt on their true reasoning capabilities. However, most proposed methods use indirect strategies or strong assumptions on pairs of questions and answers to enforce model consistency. Instead, we propose a novel strategy intended to improve model performance by directly reducing logical inconsistencies. To do this, we introduce a new consistency loss term that can be used by a wide range of the VQA models and which relies on knowing the logical relation between pairs of questions and answers. While such information is typically not available in VQA datasets, we propose to infer these logical relations using a dedicated language model and use these in our proposed consistency loss function. We conduct extensive experiments on the VQA Introspect and DME datasets and show that our method brings improvements to state-of-the-art VQA models, while being robust across different architectures and settings.
We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4 exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. A core component of this project was developing infrastructure and optimization methods that behave predictably across a wide range of scales. This allowed us to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4.
For glaciologists, studying ice sheets from the polar regions is critical. With the advancement of deep learning techniques, we can now extract high-level information from the ice sheet data (e.g., estimating the ice layer thickness, predicting the ice accumulation for upcoming years, etc.). However, a vision-based conversational deep learning approach has not been explored yet, where scientists can get information by asking questions about images. In this paper, we have introduced the task of Visual Question Answering (VQA) on remote-sensed ice sheet imagery. To study, we have presented a unique VQA dataset, Polar-VQA, in this study. All the images in this dataset were collected using four types of airborne radars. The main objective of this research is to highlight the importance of VQA in the context of ice sheet research and conduct a baseline study of existing VQA approaches on Polar-VQA dataset.