Frozen in Time: A Joint Video and Image Encoder for End-to-End Retrieval
Open accessAbstract
Our objective in this work is video-text retrieval – in particular a joint embedding that enables efficient text-to-video retrieval. The challenges in this area include the design of the visual architecture and the nature of the training data, in that the available large scale video-text training datasets, such as HowTo100M, are noisy and hence competitive performance is achieved only at scale through large amounts of compute.We address both these challenges in this paper. We propose an end-to-end trainable model that is designed to take advantage of both large-scale image and video captioning datasets. Our model is an adaptation and extension of the recent ViT and Timesformer architectures, and consists of attention in both space and time. The model is flexible and can be trained on both image and video text datasets, either independently or in conjunction. It is trained with a curriculum learning schedule that begins by treating images as ‘frozen’ snapshots of video, and then gradually learns to attend to increasing temporal context when trained on video datasets. We also provide a new video-text pretraining dataset WebVid-2M, comprised of over two million videos with weak captions scraped from the internet. Despite training on datasets that are an order of magnitude smaller, we show that this approach yields state-of-the-art results on standard downstream video-retrieval benchmarks including MSR-VTT, MSVD, DiDeMo and LSMDC.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The Platformers dataset is within an order of magnitude of other popular Internet video datasets.
“within an order of magnitude of other popular Internet video datasets (Bain et al., 2021; Wang et al., 2023).”
From Genie: Generative Interactive Environments (2024) · cites this paper as Bain et al., 2021
WebVid-10M consists of 10.7 million video clips with a total length of about 52,000 hours.
“It consists of 10.7 million video clips with a total length of about 52,000 hours.”
From Video Diffusion Models: A Survey (2024) · cites this paper as Bain et al., 2021
Cites
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 48,293 · see citing works
Wang, X., Girshick, R., Gupta, A., & He, K. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00813 · Cited by 11,596 · see citing works
Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset
Carreira, J., & Zisserman, A. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.502 · Cited by 9,570 · see citing works
Self-citationVisual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations
Krishna, R., Zhu, Y., Groth, O., Johnson, J., Hata, K., Kravitz, J., Chen, S., Kalantidis, Y., Li, L.-J., Shamma, D. A., et al. (2017) · International Journal of Computer Vision · doi:10.1007/s11263-016-0981-7 · Cited by 5,107 · see citing works
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
Sanh, V., Debut, L., Chaumond, J., & Wolf, T. (2019) · arXiv · doi:10.48550/arxiv.1910.01108 · Cited by 4,653 · see citing works
VQA: Visual Question Answering
Antol, S., Agrawal, A., Lu, J., Mitchell, M., Batra, D., Zitnick, C. L., & Parikh, D. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.279 · Cited by 4,487 · see citing works
Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision
Jia, C., Yang, Y., Xia, Y., Chen, Y.-T., Parekh, Z., Pham, H., Le, Q. V., Sung, Y., Li, Z., & Duerig, T. (2021) · arXiv · doi:10.48550/arxiv.2102.05918 · Cited by 1,977 · see citing works
Is Space-Time Attention All You Need for Video Understanding?
Bertasius, G., Wang, H., & Torresani, L. (2021) · arXiv · doi:10.48550/arxiv.2102.05095 · Cited by 1,663 · see citing works
HowTo100M: Learning a Text-Video Embedding by Watching Hundred Million Narrated Video Clips
Miech, A., Zhukov, D., Alayrac, J.-B., Tapaswi, M., Laptev, I., & Sivic, J. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00272 · Cited by 986 · see citing works
McCarthy, P. M., & Jarvis, S. (2010) · Behavior Research Methods · doi:10.3758/brm.42.2.381 · Cited by 905 · see citing works
TVQA: Localized, Compositional Video Question Answering
Lei, J., Yu, L., Bansal, M., & Berg, T. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d18-1167 · Cited by 541 · cited 1× · see citing works
“Joint visual-text models have become increasingly popular as they enable a wide suite of downstream tasks, including text-to-visual retrieval [29, 31, 36, 59], visual captioning [24, 58, 66], and visual question and answering [4, 27].”
BackgroundHERO: Hierarchical Encoder for Video+Language Omni-representation Pre-training
Li, L., Chen, Y.-C., Cheng, Y., Gan, Z., Yu, L., & Liu, J. (2020) · Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.161 · Cited by 408 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Flamingo: a visual language model for few-shot learning
Alayrac, J.-B., Donahue, J., Luc, P., Miech, A., Barr, I., Hasson, Y., Lenc, K., Mensch, A., Millican, K., Reynolds, M., et al. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-1723 · Cited by 966 · see citing works
“Another family of vision-language models is based on contrastive learning [2, 5, 49, 50, 57, 74, 82, 85, 138, 140, 146].”
BackgroundSelf-citationMovieChat: From Dense Token to Sparse Memory for Long Video Understanding
Song, E., Chai, W., Wang, G., Zhang, Y., Zhou, H., Wu, F., Chi, H., Guo, X., Ye, T., Zhang, Y., et al. (2023) · arXiv · doi:10.48550/arxiv.2307.16449 · Cited by 176 · see citing works
Phenaki: Variable Length Video Generation From Open Domain Textual Description
Villegas, R., Babaeizadeh, M., Kindermans, P.-J., Moraldo, H., Zhang, H., Saffar, M. T., Castro, S., Kunze, J., & Erhan, D. (2022) · arXiv · doi:10.48550/arxiv.2210.02399 · Cited by 92 · see citing works
“For image generation, there are datasets with billions of image-text pairs (such as LAION-5B [41] and JFT4B [60]) while the text-video datasets are substantially smaller e.g. WebVid [4] with ∼ 10 M videos, which is not enough given the higher complexity of open domain videos.”
BackgroundGenie: Generative Interactive Environments
Bruce, J., Dennis, M., Edwards, A., Parker-Holder, J., Shi, Y., Hughes, E., Lai, M., Mavalankar, A., Steigerwald, R., Apps, C., et al. (2024) · arXiv · doi:10.48550/arxiv.2402.15391 · Cited by 23 · see citing works
“The final dataset contains 6.8M 16s video clips (30k hours), within an order of magnitude of other popular Internet video datasets (Bain et al., 2021; Wang et al., 2023).”
BackgroundVideo Diffusion Models: A Survey
Melnik, A., Ljubljanac, M., Lu, C., Yan, Q., Ren, W., & Ritter, H. (2024) · arXiv · doi:10.48550/arxiv.2405.03150 · Cited by 4 · see citing works
- Authors
- Bain, M., Nagrani, A., Varol, G., & Zisserman, A.
- Affiliations
- University of Oxford (GB)
- Google (United States) (US)
- University of Oxford,Visual Geometry Group
- Venue
- 2021 IEEE/CVF International Conference on Computer Vision (ICCV)
- Publisher
- IEEE
- Published
- 2021-10
- Pages
- 1708-1718
- OpenAlex
- W3150558292
- Cited by
- 76
- References
- 12
- Citations by year
- 2026: 28 · 2025: 16 · 2024: 19 · 2023: 8 · 2022: 4 · 2021: 1
- FWCI
- 2.62
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE/CVF International Conference on Computer Vision (ICCV) (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- HAL (submittedVersion)
- arXiv (Cornell University) (cc-by)
- TIB Data Manager
- Funders
- Engineering and Physical Sciences Research Council (EP/T028572/1, VisualAI EP/T028572/1)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Video Analysis and Summarization
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image Retrieval and Classification Techniques
- Keywords
- video-text retrieval, curriculum learning, temporal attention, ViT, TimeSformer
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
