Visual Reasoning

The RefCOCO dataset is a referring expression generation (REG) dataset used for tasks related to understanding natural language expressions that refer to specific objects in images. Here are the key details about RefCOCO: 1. Collection Method: The dataset was collected using the ReferitGame, a two-player game. In this game, the first player views an image with a segmented target object and writes a natural language expression referring to that object. The second player sees only the image and the referring expression and must click on the corresponding object. If both players perform correctly, they earn points and switch roles; otherwise, they receive a new object and image for description. 2. Dataset Variants: RefCOCO: Contains 142,209 refer expressions for 50,000 objects across 19,994 images. RefCOCO+: Includes 141,564 expressions for 49,856 objects in 19,992 images. RefCOCOg: This variant has 25,799 images, 95,010 referring expressions, and 49,822 object instances. 3. Language and Restrictions: RefCOCO allows any type of language in the referring expressions. RefCOCO+ disallows location words in expressions to focus purely on appearance-based descriptions (e.g., "the man in the yellow polka-dotted shirt") rather than viewer-dependent descriptions (e.g., "the second man from the left"). These datasets serve as valuable resources for tasks like referring expression segmentation, comprehension, and visual grounding in computer vision research.

Science Question Answering (ScienceQA) is a new benchmark that consists of 21,208 multimodal multiple choice questions with diverse science topics and annotations of their answers with corresponding lectures and explanations. Out of the questions in ScienceQA, 10,332 (48.7%) have an image context, 10,220 (48.2%) have a text context, and 6,532 (30.8%) have both. Most questions are annotated with grounded lectures (83.9%) and detailed explanations (90.5%). The lecture and explanation provide general external knowledge and specific reasons, respectively, for arriving at the correct answer. To the best of our knowledge, ScienceQA is the first large-scale multimodal dataset that annotates lectures and explanations for the answers. ScienceQA, in contrast to previous datasets, has richer domain diversity from three subjects: natural science, language science, and social science. Questions in each subject are categorized first by the topic (Biology, Physics, Chemistry, etc.), then by the category (Plants, Cells, Animals, etc.), and finally by the skill (Classify fruits and vegetables as plant parts, Identify countries of Africa, etc.). ScienceQA features 26 topics, 127 categories, and 379 skills that cover a wide range of domains.

MathVista is a consolidated Mathematical reasoning benchmark within Visual contexts. It consists of three newly created datasets, IQTest, FunctionQA, and PaperQA, which address the missing visual domains and are tailored to evaluate logical reasoning on puzzle test figures, algebraic reasoning over functional plots, and scientific reasoning with academic paper figures, respectively. It also incorporates 9 MathQA datasets and 19 VQA datasets from the literature, which significantly enrich the diversity and complexity of visual perception and mathematical reasoning challenges within our benchmark. In total, MathVista includes 6,141 examples collected from 31 different datasets. - Project: https://mathvista.github.io/ - Visualization: https://mathvista.github.io/#visualization - Leaderboard: https://mathvista.github.io/#leaderboard - Paper: https://arxiv.org/abs/2310.02255 - Data: https://huggingface.co/datasets/AI4Math/MathVista - Code: https://github.com/lupantech/MathVista

Visual Commonsense Reasoning (VCR) is a large-scale dataset for cognition-level visual understanding. Given a challenging question about an image, machines need to present two sub-tasks: answer correctly and provide a rationale justifying its answer. The VCR dataset contains over 212K (training), 26K (validation) and 25K (testing) questions, answers and rationales derived from 110K movie scenes. Source: Visual Commonsense R-CNN Image Source: From Recognition to Cognition: Visual Commonsense Reasoning

SHAPES is a dataset of synthetic images designed to benchmark systems for understanding of spatial and logical relations among multiple objects. The dataset consists of complex questions about arrangements of colored shapes. The questions are built around compositions of concepts and relations, e.g. Is there a red shape above a circle? or Is a red shape blue?. Questions contain between two and four attributes, object types, or relationships. There are 244 questions and 15,616 images in total, with all questions having a yes and no answer (and corresponding supporting image). This eliminates the risk of learning biases. Each image is a 30×30 RGB image depicting a 3×3 grid of objects. Each object is characterized by shape (circle, square, triangle), colour (red, green, blue) and size (small, big). Source: Visual Question Answering: A Survey of Methods and Datasets Image Source: https://github.com/ronghanghu/n2nmn#train-and-evaluate-on-the-shapes-dataset

Visual Entailment (VE) consists 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. SNLI-VE is a dataset for VE which is based on the Stanford Natural Language Inference corpus and Flickr30k dataset. Source: https://github.com/necla-ml/SNLI-VE

Contains 145k captions for 28k images. The dataset challenges a model to recognize text, relate it to its visual context, and decide what part of the text to copy or paraphrase, requiring spatial, semantic, and visual reasoning between multiple text tokens and visual entities, such as objects. Source: TextCaps: a Dataset for Image Captioning with Reading Comprehension

RAVEN consists of 1,120,000 images and 70,000 RPM (Raven's Progressive Matrices) problems, equally distributed in 7 distinct figure configurations.

Winoground is a dataset for evaluating the ability of vision and language models to conduct visio-linguistic compositional reasoning. 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.

AbstractReasoning is a dataset for abstract reasoning, where the goal is to infer the correct answer from the context panels based on abstract reasoning. Image Source: Barrett et al

NLVR contains 92,244 pairs of human-written English sentences grounded in synthetic images. Because the images are synthetically generated, this dataset can be used for semantic parsing. Source: http://lil.nlp.cornell.edu/nlvr/ Image Source: http://lil.nlp.cornell.edu/nlvr/

The Visual Spatial Reasoning (VSR) corpus is a collection of caption-image pairs with true/false labels. Each caption describes the spatial relation of two individual objects in the image, and a vision-language model (VLM) needs to judge whether the caption is correctly describing the image (True) or not (False).

FigureQA is a visual reasoning corpus of over one million question-answer pairs grounded in over 100,000 images. The images are synthetic, scientific-style figures from five classes: line plots, dot-line plots, vertical and horizontal bar graphs, and pie charts. Source: FigureQA: An Annotated Figure Dataset for Visual Reasoning

PGM dataset serves as a tool for studying both abstract reasoning and generalisation in models. Generalisation is a multi-faceted phenomenon; there is no single, objective way in which models can or should generalise beyond their experience. The PGM dataset provides a means to measure the generalization ability of models in different ways, each of which may be more or less interesting to researchers depending on their intended training setup and applications. Source: Measuring abstract reasoning in neural networks Image Source: Barrett et al

Rendered synthetically using a library of standard 3D objects, and tests the ability to recognize compositions of object movements that require long-term reasoning. Source: CATER: A diagnostic dataset for Compositional Actions and TEmporal Reasoning Image Source: CATER

Current visual question answering (VQA) tasks mainly consider answering human-annotated questions for natural images in the daily-life context. Icon question answering (IconQA) is a benchmark which aims to highlight the importance of abstract diagram understanding and comprehensive cognitive reasoning in real-world diagram word problems. For this benchmark, a large-scale IconQA dataset is built that consists of three sub-tasks: multi-image-choice, multi-text-choice, and filling-in-the-blank. Compared to existing VQA benchmarks, IconQA requires not only perception skills like object recognition and text understanding, but also diverse cognitive reasoning skills, such as geometric reasoning, commonsense reasoning, and arithmetic reasoning. Description from: IconQA

Benchmark for physical reasoning that contains a set of simple classical mechanics puzzles in a 2D physical environment. The benchmark is designed to encourage the development of learning algorithms that are sample-efficient and generalize well across puzzles. Source: PHYRE: A New Benchmark for Physical Reasoning

The Image-Grounded Language Understanding Evaluation (IGLUE) benchmark brings together—by both aggregating pre-existing datasets and creating new ones—visual question answering, cross-modal retrieval, grounded reasoning, and grounded entailment tasks across 20 diverse languages. The benchmark enables the evaluation of multilingual multimodal models for transfer learning, not only in a zero-shot setting, but also in newly defined few-shot learning setups.

Multicultural Reasoning over Vision and Language (MaRVL) is a dataset based on an ImageNet-style hierarchy representative of many languages and cultures (Indonesian, Mandarin Chinese, Swahili, Tamil, and Turkish). The selection of both concepts and images is entirely driven by native speakers. Afterwards, we elicit statements from native speakers about pairs of images. The task consists in discriminating whether each grounded statement is true or false.

Social-IQ is an unconstrained benchmark specifically designed to train and evaluate socially intelligent technologies. By providing a rich source of open-ended questions and answers, Social-IQ opens the door to explainable social intelligence. The dataset contains rigorously annotated and validated videos, questions and answers, as well as annotations for the complexity level of each question and answer. Social-IQ contains 1,250 natural in-the-wild social situations, 7,500 questions and 52,500 correct and incorrect answers. Source: Social-IQ: A Question Answering Benchmark for Artificial Social Intelligence

Multi-modal Large Language Models (MLLMs) are increasingly prominent in the field of artificial intelligence. Although many benchmarks attempt to holistically evaluate MLLMs, they typically concentrate on basic reasoning tasks, often yielding only simple yes/no or multi-choice responses. These methods naturally lead to confusion and difficulties in conclusively determining the reasoning capabilities of MLLMs. To mitigate this issue, we manually curate CORE-MM benchmark dataset, specifically designed for MLLMs with a focus on complex reasoning tasks. Our benchmark comprises three key reasoning categories: deductive, abductive, and analogical reasoning. The queries in our dataset are intentionally constructed to engage the reasoning capabilities of MLLMs in the process of generating answers. For a fair comparison across various MLLMs, we incorporate intermediate reasoning steps into our evaluation criteria. CORE-MM benchmark consists of 279 manually curated reasoning questions, associated with a total of 342 images. Among which, 49 questions pertain to abductive reasoning, 181 require deductive reasoning, and 49 involve analogicalreasoning. Furthermore, the dataset is divided into two folds based on reasoning complexity, with 108 classified as “High” reasoning complexity and 171 as “Moderate” reasoning complexity.

CLEVR-Ref+ is a synthetic diagnostic dataset for referring expression comprehension. The precise locations and attributes of the objects are readily available, and the referring expressions are automatically associated with functional programs. The synthetic nature allows control over dataset bias (through sampling strategy), and the modular programs enable intermediate reasoning ground truth without human annotators. Source: CLEVR-Ref+: Diagnosing Visual Reasoning with Referring Expressions

WHOOPS! Is a dataset and benchmark for visual commonsense. The dataset is comprised of purposefully commonsense-defying images created by designers using publicly-available image generation tools like Midjourney. It contains commonsense-defying image from a wide range of reasons, deviations from expected social norms and everyday knowledge.

Super-CLEVR is a dataset for Visual Question Answering (VQA) where different factors in VQA domain shifts can be isolated in order that their effects can be studied independently. It contains 21 vehicle models belonging to 5 categories, with controllable attributes. Four factors are considered: visual complexity, question redundancy, concept distribution and concept compositionality. Source: Super-CLEVR: A Virtual Benchmark to Diagnose Domain Robustness in Visual Reasoning Image Source: https://arxiv.org/pdf/2212.00259v1.pdf

Cops-Ref is a dataset for visual reasoning in context of referring expression comprehension with two main features. Source: Cops-Ref: A new Dataset and Task on Compositional Referring Expression Comprehension Image Source: https://github.com/zfchenUnique/Cops-Ref

A configurable visual question and answer dataset (COG) to parallel experiments in humans and animals. COG is much simpler than the general problem of video analysis, yet it addresses many of the problems relating to visual and logical reasoning and memory -- problems that remain challenging for modern deep learning architectures. Source: A Dataset and Architecture for Visual Reasoning with a Working Memory

Geo-Diverse Visual Commonsense Reasoning (GD-VCR) is a new dataset to test vision-and-language models' ability to understand cultural and geo-location-specific commonsense. Image source: https://arxiv.org/pdf/2109.06860v1.pdf

The Relative Size dataset contains 486 object pairs between 41 physical objects. Size comparisons are not available for all pairs of objects (e.g. bird and watermelon) because for some pairs humans cannot determine which object is bigger. Dataset contains only object pairs that people have consistently agreed which one is bigger. Objects appear in 24 pairs on average, window with 13 pairs has the least, and eye with 35 pairs has the most number of comparisons. Source: Are Elephants Bigger than Butterflies? Reasoning about Sizes of Objects

Recent times have witnessed an increasing number of applications of deep neural networks towards solving tasks that require superior cognitive abilities, e.g., playing Go, generating art, ChatGPT, etc. Such a dramatic progress raises the question: how generalizable are neural networks in solving problems that demand broad skills? To answer this question, we propose SMART: a Simple Multimodal Algorithmic Reasoning Task (and the associated SMART-101 dataset) for evaluating the abstraction, deduction, and generalization abilities of neural networks in solving visuo-linguistic puzzles designed specifically for children of younger age (6--8). Our dataset consists of 101 unique puzzles; each puzzle comprises a picture and a question, and their solution needs a mix of several elementary skills, including pattern recognition, algebra, and spatial reasoning, among others. To train deep neural networks, we programmatically augment each puzzle to 2,000 new instances; each instance varied in appearance, associated natural language question, and its solution. To foster research and make progress in the quest for artificial general intelligence, we are publicly releasing our SMART-101 dataset, consisting of the full set of programmatically-generated instances of 101 puzzles and their solutions. The dataset was introduced in our paper Are Deep Neural Networks SMARTer than Second Graders? by Anoop Cherian, Kuan-Chuan Peng, Suhas Lohit, Kevin A. Smith, and Joshua B. Tenenbaum, CVPR 2023

**Compositional Physical Reasoning is a dataset for understanding object-centric and relational physics properties hidden from visual appearances. For a given set of objects, the dataset includes few videos of them moving and interacting under different initial conditions. The model is evaluated based on its capability to unravel the compositional hidden properties, such as mass and charge, and use this knowledge to answer a set of questions posted on one of the videos.

KiloGram is a resource for studying abstract visual reasoning in humans and machines. It contains a richly annotated dataset with >1k distinct stimuli. Source: Abstract Visual Reasoning with Tangram Shapes Image Source: https://arxiv.org/pdf/2211.16492v1.pdf

PGDP5K is a dataset consisting of 5000 diagram samples composed of 16 shapes, covering 5 positional relations, 22 symbol types and 6 text types, labeled with more fine-grained annotations at primitive level, including primitive classes, locations and relationships, where 1,813 non-duplicated images are selected from the Geometry3K dataset and other 3,187 images are collected from three popular textbooks across grades 6-12 on mathematics curriculum websites by taking screenshots from PDF books.

TRANCE extends CLEVR by asking a uniform question, i.e. what is the transformation between two given images, to test the ability of transformation reasoning. TRANCE includes three levels of settings, i.e. Basic (single-step transformation), Event (multi-step transformation), and View (multi-step transformation with variant views). Detailed information can be found in https://hongxin2019.github.io/TVR.

This dataset is collected via the WinoGAViL game to collect challenging vision-and-language associations. Inspired by the popular card game Codenames, a “spymaster” gives a textual cue related to several visual candidates, and another player has to identify them. We use the game to collect 3.5K instances, finding that they are intuitive for humans (>90% Jaccard index) but challenging for state-of-the-art AI models, where the best model (ViLT) achieves a score of 52%, succeeding mostly where the cue is visually salient. Researchers are welcome to evaluate models on this dataset. A simple intended use is zero-shot prediction: run vision-and-language model, producing a score for the (cue,image) pair, and taking the K pairs with the highest scores. A supervised setting is also possible, code for re-running the experiments is available in the github repository. https://github.com/WinoGAViL/WinoGAViL-experiments

Visual Analogies of Situation Recognition (VASR) is a dataset for visual analogical mapping, adapting the classical word-analogy task into the visual domain. It contains 196K object transitions and 385K activity transitions. Experiments demonstrate that state-of-the-art models do well when distractors are chosen randomly (~86%), but struggle with carefully chosen distractors (~53%, compared to 90% human accuracy) Source: VASR: Visual Analogies of Situation Recognition Image Source:https://arxiv.org/pdf/2212.04542v1.pdf Project Website: https://vasr-dataset.github.io/

ADE-Affordance is a new dataset that builds upon ADE20k, which contains annotations enabling such rich visual reasoning. Source: Learning to Act Properly: Predicting and Explaining Affordances from Images