{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/visual-entailment-a-novel-task-for-fine","title":"Visual Entailment: A Novel Task for Fine-Grained Image Understanding","arxiv_id":"1901.06706","date":"2019-01-20","proceeding":null,"authors":["Ning Xie","Farley Lai","Derek Doran","Asim Kadav"],"abstract":"Existing visual reasoning datasets such as Visual Question Answering (VQA),\noften suffer from biases conditioned on the question, image or answer\ndistributions. The recently proposed CLEVR dataset addresses these limitations\nand requires fine-grained reasoning but the dataset is synthetic and consists\nof similar objects and sentence structures across the dataset.\n  In this paper, we introduce a new inference task, Visual Entailment (VE) -\nconsisting of image-sentence pairs whereby a premise is defined by an image,\nrather than a natural language sentence as in traditional Textual Entailment\ntasks. The goal of a trained VE model is to predict whether the image\nsemantically entails the text. To realize this task, we build a dataset SNLI-VE\nbased on the Stanford Natural Language Inference corpus and Flickr30k dataset.\nWe evaluate various existing VQA baselines and build a model called Explainable\nVisual Entailment (EVE) system to address the VE task. EVE achieves up to 71%\naccuracy and outperforms several other state-of-the-art VQA based models.\nFinally, we demonstrate the explainability of EVE through cross-modal attention\nvisualizations. The SNLI-VE dataset is publicly available at\nhttps://github.com/ necla-ml/SNLI-VE.","url_abs":"http://arxiv.org/abs/1901.06706v1","url_pdf":"http://arxiv.org/pdf/1901.06706v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"visual-entailment-a-novel-task-for-fine","repo_url":"https://github.com/necla-ml/SNLI-VE","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"BSD-3-Clause"}}],"tasks":[{"task_slug":"natural-language-inference","task_name":"Natural Language Inference"},{"task_slug":"question-answering","task_name":"Question Answering"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"visual-entailment","task_name":"Visual Entailment"},{"task_slug":"visual-question-answering-1","task_name":"Visual Question Answering"},{"task_slug":"visual-question-answering","task_name":"Visual Question Answering (VQA)"},{"task_slug":"visual-reasoning","task_name":"Visual Reasoning"}],"methods":[],"datasets_introduced":[{"slug":"snli-ve","name":"SNLI-VE","full_name":null}],"methods_introduced":[],"results":[{"leaderboard":"/sota/visual-entailment-on-snli-ve-test","task":"Visual Entailment","dataset":"SNLI-VE test","model":"EVE-ROI*","rank_in_archive_order":8,"of":8,"metrics":{"Accuracy":"70.47"},"uses_additional_data":false},{"leaderboard":"/sota/visual-entailment-on-snli-ve-val","task":"Visual Entailment","dataset":"SNLI-VE val","model":"EVE-ROI*","rank_in_archive_order":9,"of":9,"metrics":{"Accuracy":"70.81"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1901.06706","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}