# Memory Warps for Learning Long-Term Online Video Representations

## TL;DR

This paper introduces a memory-based online video representation that efficiently captures long-term information, aligns features over time, and predicts future frames, significantly improving real-time object detection performance.

## Contribution

It presents a novel memory mechanism that is spatially warped over time, enabling long-term, predictive video representations in an online setting, unlike prior offline or computationally heavy methods.

## Key findings

- Achieved 2.3x speed-up in object detection on ImageNet-VID.
- Lost only 0.9% mAP compared to offline methods.
- Enhanced real-time detection by over 10% mAP using feature propagation.

## Abstract

This paper proposes a novel memory-based online video representation that is efficient, accurate and predictive. This is in contrast to prior works that often rely on computationally heavy 3D convolutions, ignore actual motion when aligning features over time, or operate in an off-line mode to utilize future frames. In particular, our memory (i) holds the feature representation, (ii) is spatially warped over time to compensate for observer and scene motions, (iii) can carry long-term information, and (iv) enables predicting feature representations in future frames. By exploring a variant that operates at multiple temporal scales, we efficiently learn across even longer time horizons. We apply our online framework to object detection in videos, obtaining a large 2.3 times speed-up and losing only 0.9% mAP on ImageNet-VID dataset, compared to prior works that even use future frames. Finally, we demonstrate the predictive property of our representation in two novel detection setups, where features are propagated over time to (i) significantly enhance a real-time detector by more than 10% mAP in a multi-threaded online setup and to (ii) anticipate objects in future frames.