Representation Recycling for Streaming Video Analysis

TL;DR

StreamDEQ introduces a streaming video analysis method that recycles inference computations across frames, significantly reducing processing time while maintaining high accuracy in tasks like segmentation, detection, and pose estimation.

Contribution

The paper proposes a novel streaming recognition model using implicit layers that recycles inference iterations across video frames to improve efficiency.

Findings

Achieves 2-4x faster processing than baseline methods.

Maintains comparable accuracy in video segmentation, detection, and pose estimation.

Effectively recovers near-optimal representations with minimal frames.

Abstract

We present StreamDEQ, a method that aims to infer frame-wise representations on videos with minimal per-frame computation. Conventional deep networks do feature extraction from scratch at each frame in the absence of ad-hoc solutions. We instead aim to build streaming recognition models that can natively exploit temporal smoothness between consecutive video frames. We observe that the recently emerging implicit layer models provide a convenient foundation to construct such models, as they define representations as the fixed-points of shallow networks, which need to be estimated using iterative methods. Our main insight is to distribute the inference iterations over the temporal axis by using the most recent representation as a starting point at each frame. This scheme effectively recycles the recent inference computations and greatly reduces the needed processing time. Through extensive experimental analysis, we show that StreamDEQ is able to recover near-optimal representations in a few frames' time and maintain an up-to-date representation throughout the video duration. Our experiments on video semantic segmentation, video object detection, and human pose estimation in videos show that StreamDEQ achieves on-par accuracy with the baseline while being more than 2-4x faster.