We present a novel task and dataset for evaluating the ability of vision and language models to conduct visio-linguistic compositional reasoning, which we call Winoground. Given two images and two captions, the goal is to match them correctly - but crucially, both captions contain a completely identical set of words, only in a different order. The dataset was carefully hand-curated by expert annotators and is labeled with a rich set of fine-grained tags to assist in analyzing model performance. We probe a diverse range of state-of-the-art vision and language models and find that, surprisingly, none of them do much better than chance. Evidently, these models are not as skilled at visio-linguistic compositional reasoning as we might have hoped. We perform an extensive analysis to obtain insights into how future work might try to mitigate these models' shortcomings. We aim for Winoground to serve as a useful evaluation set for advancing the state of the art and driving further progress in the field. The dataset is available at https://huggingface.co/datasets/facebook/winoground.
Providing explanations in the context of Visual Question Answering (VQA) presents a fundamental problem in machine learning. To obtain detailed insights into the process of generating natural language explanations for VQA, we introduce the large-scale CLEVR-X dataset that extends the CLEVR dataset with natural language explanations. For each image-question pair in the CLEVR dataset, CLEVR-X contains multiple structured textual explanations which are derived from the original scene graphs. By construction, the CLEVR-X explanations are correct and describe the reasoning and visual information that is necessary to answer a given question. We conducted a user study to confirm that the ground-truth explanations in our proposed dataset are indeed complete and relevant. We present baseline results for generating natural language explanations in the context of VQA using two state-of-the-art frameworks on the CLEVR-X dataset. Furthermore, we provide a detailed analysis of the explanation generation quality for different question and answer types. Additionally, we study the influence of using different numbers of ground-truth explanations on the convergence of natural language generation (NLG) metrics. The CLEVR-X dataset is publicly available at \url{https://explainableml.github.io/CLEVR-X/}.
Most existing approaches to Visual Question Answering (VQA) answer questions directly, however, people usually decompose a complex question into a sequence of simple sub questions and finally obtain the answer to the original question after answering the sub question sequence(SQS). By simulating the process, this paper proposes a conversation-based VQA (Co-VQA) framework, which consists of three components: Questioner, Oracle, and Answerer. Questioner raises the sub questions using an extending HRED model, and Oracle answers them one-by-one. An Adaptive Chain Visual Reasoning Model (ACVRM) for Answerer is also proposed, where the question-answer pair is used to update the visual representation sequentially. To perform supervised learning for each model, we introduce a well-designed method to build a SQS for each question on VQA 2.0 and VQA-CP v2 datasets. Experimental results show that our method achieves state-of-the-art on VQA-CP v2. Further analyses show that SQSs help build direct semantic connections between questions and images, provide question-adaptive variable-length reasoning chains, and with explicit interpretability as well as error traceability.
Grounded situation recognition is the task of predicting the main activity, entities playing certain roles within the activity, and bounding-box groundings of the entities in the given image. To effectively deal with this challenging task, we introduce a novel approach where the two processes for activity classification and entity estimation are interactive and complementary. To implement this idea, we propose Collaborative Glance-Gaze TransFormer (CoFormer) that consists of two modules: Glance transformer for activity classification and Gaze transformer for entity estimation. Glance transformer predicts the main activity with the help of Gaze transformer that analyzes entities and their relations, while Gaze transformer estimates the grounded entities by focusing only on the entities relevant to the activity predicted by Glance transformer. Our CoFormer achieves the state of the art in all evaluation metrics on the SWiG dataset. Training code and model weights are available at https://github.com/jhcho99/CoFormer.
Effectiveness and interpretability are two essential properties for trustworthy AI systems. Most recent studies in visual reasoning are dedicated to improving the accuracy of predicted answers, and less attention is paid to explaining the rationales behind the decisions. As a result, they commonly take advantage of spurious biases instead of actually reasoning on the visual-textual data, and have yet developed the capability to explain their decision making by considering key information from both modalities. This paper aims to close the gap from three distinct perspectives: first, we define a new type of multi-modal explanations that explain the decisions by progressively traversing the reasoning process and grounding keywords in the images. We develop a functional program to sequentially execute different reasoning steps and construct a new dataset with 1,040,830 multi-modal explanations. Second, we identify the critical need to tightly couple important components across the visual and textual modalities for explaining the decisions, and propose a novel explanation generation method that explicitly models the pairwise correspondence between words and regions of interest. It improves the visual grounding capability by a considerable margin, resulting in enhanced interpretability and reasoning performance. Finally, with our new data and method, we perform extensive analyses to study the effectiveness of our explanation under different settings, including multi-task learning and transfer learning. Our code and data are available at https://github.com/szzexpoi/rex.
Current state-of-the-art one-stage object detectors are limited by treating each image region separately without considering possible relations of the objects. This causes dependency solely on high-quality convolutional feature representations for detecting objects successfully. However, this may not be possible sometimes due to some challenging conditions. In this paper, the usage of reasoning features on one-stage object detection is analyzed. We attempted different architectures that reason the relations of the image regions by using self-attention. YOLOv3-Reasoner2 model spatially and semantically enhances features in the reasoning layer and fuses them with the original convolutional features to improve performance. The YOLOv3-Reasoner2 model achieves around 2.5% absolute improvement with respect to baseline YOLOv3 on COCO in terms of mAP while still running in real-time.
Visual Commonsense Reasoning (VCR), deemed as one challenging extension of the Visual Question Answering (VQA), endeavors to pursue a more high-level visual comprehension. It is composed of two indispensable processes: question answering over a given image and rationale inference for answer explanation. Over the years, a variety of methods tackling VCR have advanced the performance on the benchmark dataset. Despite significant as these methods are, they often treat the two processes in a separate manner and hence decompose the VCR into two irrelevant VQA instances. As a result, the pivotal connection between question answering and rationale inference is interrupted, rendering existing efforts less faithful on visual reasoning. To empirically study this issue, we perform some in-depth explorations in terms of both language shortcuts and generalization capability to verify the pitfalls of this treatment. Based on our findings, in this paper, we present a plug-and-play knowledge distillation enhanced framework to couple the question answering and rationale inference processes. The key contribution is the introduction of a novel branch, which serves as the bridge to conduct processes connecting. Given that our framework is model-agnostic, we apply it to the existing popular baselines and validate its effectiveness on the benchmark dataset. As detailed in the experimental results, when equipped with our framework, these baselines achieve consistent and significant performance improvements, demonstrating the viability of processes coupling, as well as the superiority of the proposed framework.
How can we measure the reasoning capabilities of intelligence systems? Visual question answering provides a convenient framework for testing the model's abilities by interrogating the model through questions about the scene. However, despite scores of various visual QA datasets and architectures, which sometimes yield even a super-human performance, the question of whether those architectures can actually reason remains open to debate. To answer this, we extend the visual question answering framework and propose the following behavioral test in the form of a two-player game. We consider black-box neural models of CLEVR. These models are trained on a diagnostic dataset benchmarking reasoning. Next, we train an adversarial player that re-configures the scene to fool the CLEVR model. We show that CLEVR models, which otherwise could perform at a human level, can easily be fooled by our agent. Our results put in doubt whether data-driven approaches can do reasoning without exploiting the numerous biases that are often present in those datasets. Finally, we also propose a controlled experiment measuring the efficiency of such models to learn and perform reasoning.
Abstract Visual Reasoning (AVR) problems are commonly used to approximate human intelligence. They test the ability of applying previously gained knowledge, experience and skills in a completely new setting, which makes them particularly well-suited for this task. Recently, the AVR problems have become popular as a proxy to study machine intelligence, which has led to emergence of new distinct types of problems and multiple benchmark sets. In this work we review this emerging AVR research and propose a taxonomy to categorise the AVR tasks along 5 dimensions: input shapes, hidden rules, target task, cognitive function, and main challenge. The perspective taken in this survey allows to characterise AVR problems with respect to their shared and distinct properties, provides a unified view on the existing approaches for solving AVR tasks, shows how the AVR problems relate to practical applications, and outlines promising directions for future work. One of them refers to the observation that in the machine learning literature different tasks are considered in isolation, which is in the stark contrast with the way the AVR tasks are used to measure human intelligence, where multiple types of problems are combined within a single IQ test.
We present Grammar-Based Grounded Lexicon Learning (G2L2), a lexicalist approach toward learning a compositional and grounded meaning representation of language from grounded data, such as paired images and texts. At the core of G2L2 is a collection of lexicon entries, which map each word to a tuple of a syntactic type and a neuro-symbolic semantic program. For example, the word shiny has a syntactic type of adjective; its neuro-symbolic semantic program has the symbolic form {\lambda}x. filter(x, SHINY), where the concept SHINY is associated with a neural network embedding, which will be used to classify shiny objects. Given an input sentence, G2L2 first looks up the lexicon entries associated with each token. It then derives the meaning of the sentence as an executable neuro-symbolic program by composing lexical meanings based on syntax. The recovered meaning programs can be executed on grounded inputs. To facilitate learning in an exponentially-growing compositional space, we introduce a joint parsing and expected execution algorithm, which does local marginalization over derivations to reduce the training time. We evaluate G2L2 on two domains: visual reasoning and language-driven navigation. Results show that G2L2 can generalize from small amounts of data to novel compositions of words.
Humans have remarkable capacity to reason abductively and hypothesize about what lies beyond the literal content of an image. By identifying concrete visual clues scattered throughout a scene, we almost can't help but draw probable inferences beyond the literal scene based on our everyday experience and knowledge about the world. For example, if we see a "20 mph" sign alongside a road, we might assume the street sits in a residential area (rather than on a highway), even if no houses are pictured. Can machines perform similar visual reasoning? We present Sherlock, an annotated corpus of 103K images for testing machine capacity for abductive reasoning beyond literal image contents. We adopt a free-viewing paradigm: participants first observe and identify salient clues within images (e.g., objects, actions) and then provide a plausible inference about the scene, given the clue. In total, we collect 363K (clue, inference) pairs, which form a first-of-its-kind abductive visual reasoning dataset. Using our corpus, we test three complementary axes of abductive reasoning. We evaluate the capacity of models to: i) retrieve relevant inferences from a large candidate corpus; ii) localize evidence for inferences via bounding boxes, and iii) compare plausible inferences to match human judgments on a newly-collected diagnostic corpus of 19K Likert-scale judgments. While we find that fine-tuning CLIP-RN50x64 with a multitask objective outperforms strong baselines, significant headroom exists between model performance and human agreement. Data, models, and leaderboard available at http://visualabduction.com/
Recently, DALL-E, a multimodal transformer language model, and its variants, including diffusion models, have shown high-quality text-to-image generation capabilities. However, despite the realistic image generation results, there has not been a detailed analysis of how to evaluate such models. In this work, we investigate the visual reasoning capabilities and social biases of different text-to-image models, covering both multimodal transformer language models and diffusion models. First, we measure three visual reasoning skills: object recognition, object counting, and spatial relation understanding. For this, we propose PaintSkills, a compositional diagnostic evaluation dataset that measures these skills. Despite the high-fidelity image generation capability, a large gap exists between the performance of recent models and the upper bound accuracy in object counting and spatial relation understanding skills. Second, we assess the gender and skin tone biases by measuring the gender/skin tone distribution of generated images across various professions and attributes. We demonstrate that recent text-to-image generation models learn specific biases about gender and skin tone from web image-text pairs. We hope our work will help guide future progress in improving text-to-image generation models on visual reasoning skills and learning socially unbiased representations. Code and data: https://github.com/j-min/DallEval
In this work, we pursue a unified paradigm for multimodal pretraining to break the scaffolds of complex task/modality-specific customization. We propose OFA, a Task-Agnostic and Modality-Agnostic framework that supports Task Comprehensiveness. OFA unifies a diverse set of cross-modal and unimodal tasks, including image generation, visual grounding, image captioning, image classification, language modeling, etc., in a simple sequence-to-sequence learning framework. OFA follows the instruction-based learning in both pretraining and finetuning stages, requiring no extra task-specific layers for downstream tasks. In comparison with the recent state-of-the-art vision & language models that rely on extremely large cross-modal datasets, OFA is pretrained on only 20M publicly available image-text pairs. Despite its simplicity and relatively small-scale training data, OFA achieves new SOTAs in a series of cross-modal tasks while attaining highly competitive performances on uni-modal tasks. Our further analysis indicates that OFA can also effectively transfer to unseen tasks and unseen domains. Our code and models are publicly available at https://github.com/OFA-Sys/OFA.
Vector Quantization (VQ) is a method for discretizing latent representations and has become a major part of the deep learning toolkit. It has been theoretically and empirically shown that discretization of representations leads to improved generalization, including in reinforcement learning where discretization can be used to bottleneck multi-agent communication to promote agent specialization and robustness. The discretization tightness of most VQ-based methods is defined by the number of discrete codes in the representation vector and the codebook size, which are fixed as hyperparameters. In this work, we propose learning to dynamically select discretization tightness conditioned on inputs, based on the hypothesis that data naturally contains variations in complexity that call for different levels of representational coarseness. We show that dynamically varying tightness in communication bottlenecks can improve model performance on visual reasoning and reinforcement learning tasks.
Abstract visual reasoning (AVR) domain encompasses problems solving which requires the ability to reason about relations among entities present in a given scene. While humans, generally, solve AVR tasks in a "natural" way, even without prior experience, this type of problems has proven difficult for current machine learning systems. The paper summarises recent progress in applying deep learning methods to solving AVR problems, as a proxy for studying machine intelligence. We focus on the most common type of AVR tasks -- the Raven's Progressive Matrices (RPMs) -- and provide a comprehensive review of the learning methods and deep neural models applied to solve RPMs, as well as, the RPM benchmark sets. Performance analysis of the state-of-the-art approaches to solving RPMs leads to formulation of certain insights and remarks on the current and future trends in this area. We conclude the paper by demonstrating how real-world problems can benefit from the discoveries of RPM studies.
Vision-Language Pre-training (VLP) has advanced the performance for many vision-language tasks. However, most existing pre-trained models only excel in either understanding-based tasks or generation-based tasks. Furthermore, performance improvement has been largely achieved by scaling up the dataset with noisy image-text pairs collected from the web, which is a suboptimal source of supervision. In this paper, we propose BLIP, a new VLP framework which transfers flexibly to both vision-language understanding and generation tasks. BLIP effectively utilizes the noisy web data by bootstrapping the captions, where a captioner generates synthetic captions and a filter removes the noisy ones. We achieve state-of-the-art results on a wide range of vision-language tasks, such as image-text retrieval (+2.7% in average recall@1), image captioning (+2.8% in CIDEr), and VQA (+1.6% in VQA score). BLIP also demonstrates strong generalization ability when directly transferred to video-language tasks in a zero-shot manner. Code, models, and datasets are released at https://github.com/salesforce/BLIP.
We focus on the confounding bias between language and location in the visual grounding pipeline, where we find that the bias is the major visual reasoning bottleneck. For example, the grounding process is usually a trivial language-location association without visual reasoning, e.g., grounding any language query containing sheep to the nearly central regions, due to that most queries about sheep have ground-truth locations at the image center. First, we frame the visual grounding pipeline into a causal graph, which shows the causalities among image, query, target location and underlying confounder. Through the causal graph, we know how to break the grounding bottleneck: deconfounded visual grounding. Second, to tackle the challenge that the confounder is unobserved in general, we propose a confounder-agnostic approach called: Referring Expression Deconfounder (RED), to remove the confounding bias. Third, we implement RED as a simple language attention, which can be applied in any grounding method. On popular benchmarks, RED improves various state-of-the-art grounding methods by a significant margin. Code will soon be available at: https://github.com/JianqiangH/Deconfounded_VG.
Visual Question Answering on 3D Point Cloud (VQA-3D) is an emerging yet challenging field that aims at answering various types of textual questions given an entire point cloud scene. To tackle this problem, we propose the CLEVR3D, a large-scale VQA-3D dataset consisting of 171K questions from 8,771 3D scenes. Specifically, we develop a question engine leveraging 3D scene graph structures to generate diverse reasoning questions, covering the questions of objects' attributes (i.e., size, color, and material) and their spatial relationships. Through such a manner, we initially generated 44K questions from 1,333 real-world scenes. Moreover, a more challenging setup is proposed to remove the confounding bias and adjust the context from a common-sense layout. Such a setup requires the network to achieve comprehensive visual understanding when the 3D scene is different from the general co-occurrence context (e.g., chairs always exist with tables). To this end, we further introduce the compositional scene manipulation strategy and generate 127K questions from 7,438 augmented 3D scenes, which can improve VQA-3D models for real-world comprehension. Built upon the proposed dataset, we baseline several VQA-3D models, where experimental results verify that the CLEVR3D can significantly boost other 3D scene understanding tasks. Our code and dataset will be made publicly available at https://github.com/yanx27/CLEVR3D.
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.
A critical aspect of human visual perception is the ability to parse visual scenes into individual objects and further into object parts, forming part-whole hierarchies. Such composite structures could induce a rich set of semantic concepts and relations, thus playing an important role in the interpretation and organization of visual signals as well as for the generalization of visual perception and reasoning. However, existing visual reasoning benchmarks mostly focus on objects rather than parts. Visual reasoning based on the full part-whole hierarchy is much more challenging than object-centric reasoning due to finer-grained concepts, richer geometry relations, and more complex physics. Therefore, to better serve for part-based conceptual, relational and physical reasoning, we introduce a new large-scale diagnostic visual reasoning dataset named PTR. PTR contains around 70k RGBD synthetic images with ground truth object and part level annotations regarding semantic instance segmentation, color attributes, spatial and geometric relationships, and certain physical properties such as stability. These images are paired with 700k machine-generated questions covering various types of reasoning types, making them a good testbed for visual reasoning models. We examine several state-of-the-art visual reasoning models on this dataset and observe that they still make many surprising mistakes in situations where humans can easily infer the correct answer. We believe this dataset will open up new opportunities for part-based reasoning.