Memory Consolidation Enables Long-Context Video Understanding

Ivana Balazevic, Yuge Shi, Pinelopi Papalampidi, Rahma Chaabouni, Skanda Koppula, Olivier J Henaff
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:2527-2542, 2024.

Abstract

Most transformer-based video encoders are limited to short temporal contexts due to their quadratic complexity. While various attempts have been made to extend this context, this has often come at the cost of both conceptual and computational complexity. We propose to instead re-purpose existing pre-trained video transformers by simply fine-tuning them to attend to memories derived non-parametrically from past activations. By leveraging redundancy reduction, our memory-consolidated vision transformer (MC-ViT) effortlessly extends its context far into the past and exhibits excellent scaling behavior when learning from longer videos. In doing so, MC-ViT sets a new state-of-the-art in long-context video understanding on EgoSchema, Perception Test, and Diving48, outperforming methods that benefit from orders of magnitude more parameters.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-balazevic24a, title = {Memory Consolidation Enables Long-Context Video Understanding}, author = {Balazevic, Ivana and Shi, Yuge and Papalampidi, Pinelopi and Chaabouni, Rahma and Koppula, Skanda and Henaff, Olivier J}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {2527--2542}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/balazevic24a/balazevic24a.pdf}, url = {https://proceedings.mlr.press/v235/balazevic24a.html}, abstract = {Most transformer-based video encoders are limited to short temporal contexts due to their quadratic complexity. While various attempts have been made to extend this context, this has often come at the cost of both conceptual and computational complexity. We propose to instead re-purpose existing pre-trained video transformers by simply fine-tuning them to attend to memories derived non-parametrically from past activations. By leveraging redundancy reduction, our memory-consolidated vision transformer (MC-ViT) effortlessly extends its context far into the past and exhibits excellent scaling behavior when learning from longer videos. In doing so, MC-ViT sets a new state-of-the-art in long-context video understanding on EgoSchema, Perception Test, and Diving48, outperforming methods that benefit from orders of magnitude more parameters.} }
Endnote
%0 Conference Paper %T Memory Consolidation Enables Long-Context Video Understanding %A Ivana Balazevic %A Yuge Shi %A Pinelopi Papalampidi %A Rahma Chaabouni %A Skanda Koppula %A Olivier J Henaff %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-balazevic24a %I PMLR %P 2527--2542 %U https://proceedings.mlr.press/v235/balazevic24a.html %V 235 %X Most transformer-based video encoders are limited to short temporal contexts due to their quadratic complexity. While various attempts have been made to extend this context, this has often come at the cost of both conceptual and computational complexity. We propose to instead re-purpose existing pre-trained video transformers by simply fine-tuning them to attend to memories derived non-parametrically from past activations. By leveraging redundancy reduction, our memory-consolidated vision transformer (MC-ViT) effortlessly extends its context far into the past and exhibits excellent scaling behavior when learning from longer videos. In doing so, MC-ViT sets a new state-of-the-art in long-context video understanding on EgoSchema, Perception Test, and Diving48, outperforming methods that benefit from orders of magnitude more parameters.
APA
Balazevic, I., Shi, Y., Papalampidi, P., Chaabouni, R., Koppula, S. & Henaff, O.J.. (2024). Memory Consolidation Enables Long-Context Video Understanding. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:2527-2542 Available from https://proceedings.mlr.press/v235/balazevic24a.html.

Related Material