{"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/tips-and-tricks-for-visual-question-answering","title":"Tips and Tricks for Visual Question Answering: Learnings from the 2017 Challenge","arxiv_id":"1708.02711","date":"2017-08-09","proceeding":"CVPR 2018 6","authors":["Damien Teney","Peter Anderson","Xiaodong He","Anton Van Den Hengel"],"abstract":"This paper presents a state-of-the-art model for visual question answering\n(VQA), which won the first place in the 2017 VQA Challenge. VQA is a task of\nsignificant importance for research in artificial intelligence, given its\nmultimodal nature, clear evaluation protocol, and potential real-world\napplications. The performance of deep neural networks for VQA is very dependent\non choices of architectures and hyperparameters. To help further research in\nthe area, we describe in detail our high-performing, though relatively simple\nmodel. Through a massive exploration of architectures and hyperparameters\nrepresenting more than 3,000 GPU-hours, we identified tips and tricks that lead\nto its success, namely: sigmoid outputs, soft training targets, image features\nfrom bottom-up attention, gated tanh activations, output embeddings initialized\nusing GloVe and Google Images, large mini-batches, and smart shuffling of\ntraining data. We provide a detailed analysis of their impact on performance to\nassist others in making an appropriate selection.","url_abs":"http://arxiv.org/abs/1708.02711v1","url_pdf":"http://arxiv.org/pdf/1708.02711v1.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":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/SinghJasdeep/Attention-on-Attention-for-VQA","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/VincentYing/Attention-on-Attention-for-VQA","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/feifengwhu/question_attention","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/hengyuan-hu/bottom-up-attention-vqa","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/peteanderson80/bottom-up-attention","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/shailzajolly/Understanding-yesno-and-nonyesno-samples-VQA","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/shailzajolly/icdar_vqa","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/snagiri/ECE285_Jarvis_ProjectA","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/thilinicooray/Bottom-up-vqa","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"tips-and-tricks-for-visual-question-answering","repo_url":"https://github.com/yangdsh/VQA-BUTD-demo","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":null,"task_name":"GPU"},{"task_slug":"visual-question-answering-1","task_name":"Visual Question Answering"},{"task_slug":"visual-question-answering","task_name":"Visual Question Answering (VQA)"}],"methods":[{"method_slug":"glove","method_name":"GloVe"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/visual-question-answering-on-vqa-v2-test-dev","task":"Visual Question Answering (VQA)","dataset":"VQA v2 test-dev","model":"Image features from bottom-up attention (adaptive K, ensemble)","rank_in_archive_order":33,"of":56,"metrics":{"Accuracy":"69.87"},"uses_additional_data":false},{"leaderboard":"/sota/visual-question-answering-on-vqa-v2-test-std","task":"Visual Question Answering (VQA)","dataset":"VQA v2 test-std","model":"Image features from bottom-up attention (adaptive K, ensemble)","rank_in_archive_order":29,"of":38,"metrics":{"overall":"70.3"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1708.02711","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}