{"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/late-temporal-modeling-in-3d-cnn","title":"Late Temporal Modeling in 3D CNN Architectures with BERT for Action Recognition","arxiv_id":"2008.01232","date":"2020-08-03","proceeding":null,"authors":["M. Esat Kalfaoglu","Sinan Kalkan","A. Aydin Alatan"],"abstract":"In this work, we combine 3D convolution with late temporal modeling for action recognition. For this aim, we replace the conventional Temporal Global Average Pooling (TGAP) layer at the end of 3D convolutional architecture with the Bidirectional Encoder Representations from Transformers (BERT) layer in order to better utilize the temporal information with BERT's attention mechanism. We show that this replacement improves the performances of many popular 3D convolution architectures for action recognition, including ResNeXt, I3D, SlowFast and R(2+1)D. Moreover, we provide the-state-of-the-art results on both HMDB51 and UCF101 datasets with 85.10% and 98.69% top-1 accuracy, respectively. The code is publicly available.","url_abs":"https://arxiv.org/abs/2008.01232v3","url_pdf":"https://arxiv.org/pdf/2008.01232v3.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":"late-temporal-modeling-in-3d-cnn","repo_url":"https://github.com/artest08/LateTemporalModeling3DCNN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"late-temporal-modeling-in-3d-cnn","repo_url":"https://github.com/kietngt00/hmdb51-recognition","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"action-recognition-in-videos","task_name":"Action Recognition"}],"methods":[{"method_slug":"2-1-d-convolution","method_name":"(2+1)D Convolution"},{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"3d-convolution","method_name":"3D Convolution"},{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"grouped-convolution","method_name":"Grouped Convolution"},{"method_slug":"kaiming-initialization","method_name":"Kaiming Initialization"},{"method_slug":"r-2-1-d","method_name":"R(2+1)D"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"resnext","method_name":"ResNeXt"},{"method_slug":"resnext-block","method_name":"ResNeXt Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/action-recognition-in-videos-on-hmdb-51","task":"Action Recognition","dataset":"HMDB-51","model":"R2+1D-BERT","rank_in_archive_order":7,"of":77,"metrics":{"Average accuracy of 3 splits":"85.10"},"uses_additional_data":true},{"leaderboard":"/sota/action-recognition-on-ucf-101","task":"Action Recognition","dataset":"UCF 101","model":"R2+1D-BERT","rank_in_archive_order":1,"of":1,"metrics":{"3-fold Accuracy":"98.69"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2008.01232","atlas_url":"https://app.syntology.ai/?focus=2008.01232","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}