AI Paper Atlas

Research map / Video / Video segmentation and tracking

Dynamic Temporal Filtering In Video Models

Fuchen Long, Zhaofan Qiu, Yingwei Pan, Ting Yao, Chong-Wah Ngo, Tao Mei

ECCV 2022 · Video › Video segmentation and tracking · approach: Self-supervised video representation learning · 39 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace temporal 1D convolution with dynamic 1D convolution per location (ablation)DTF-Net (res5) · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 73.2 -> 74.2Improved
move temporal modeling to frequency domain with a fixed filter (ablation)DTF-Net (res5) · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 74.2 -> 75.0Improved
make the frequency filter dynamic per spatial location (ablation)DTF-Net (res5) · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 75.0 -> 75.7Improved
add DTF block to 2D-ResNet at res5 (ablation)ResNet-50 · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 72.0 -> 75.7Improved
add frame-wise aggregation before FFT (ablation)DTF-baseline · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 74.9 -> 75.4Improved
add correlation feature to filter learning (ablation)DTF-baseline · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 74.9 -> 75.2Improved
combine frame-wise aggregation and correlation feature (ablation)DTF-baseline · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 74.9 -> 75.7Improved
insert DTF into res4 in addition to res5 (ablation)DTF-Net ResNet-50 · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 75.7 -> 76.5Improved
insert DTF into all four stages (ablation)DTF-Net ResNet-50 · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 76.5 -> 77.7Improved
use DTF instead of Temporal Difference for temporal modelingResNet-50 · Kinetics-400 val (ablation, 16 frames)top-1 accuracy: 76.6 -> 77.7Improved
increase input clip length from 16 to 64 frames (ablation)DTF-Net ResNet-50 · Kinetics-400 valtop-1 accuracy: 77.7 -> 80.9Improved
replace Video-Swin with DTF-Transformer (add DTF to Swin-B)Swin-B · Kinetics-400 valtop-1 accuracy: 82.7 -> 83.5Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy