AI Paper Atlas

Research map / Video / Video segmentation and tracking

Video Object Segmentation With Dynamic Memory Networks and Adaptive Object Alignment

Shuxian Liang, Xu Shen, Jianqiang Huang, Xian-Sheng Hua

ICCV 2021 · Video › Video segmentation and tracking · approach: Mask propagation for unsupervised video segmentation · 30 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
add DMN (dynamic memory template) to fixed first-frame template (ablation)Base model · DAVIS 2017 valG: 77.8 -> 80.5Improved
add AOA (adaptive object alignment) to base model (ablation)Base model · DAVIS 2017 valG: 77.8 -> 78.9Improved
use object features instead of frame features in memory encoder (ablation)Full model · DAVIS 2017 valG: 79.4 -> 81.3Improved
add matcher with AOA and proposals (vs decoding masks from DMN) (ablation)Full model · DAVIS 2017 valG: 70.0 -> 81.3Improved
add RefineNet on coarse masks (ablation)Full model · DAVIS 2017 valG: 69.2 -> 81.3Improved
add intermediate memory frames (stride 5) to first+previous (ablation)Full model · DAVIS 2017 valG: 80.4 -> 81.3Improved
use all past frames in memory instead of stride-5 sampling (ablation)Full model · DAVIS 2017 valG / time: 81.3 -> 81.8 (0.15 -> 0.19 s/frame)Mixed
add main training on videos after pre-training (ablation)Full model · DAVIS 2017 valG: 72.5 -> 81.3Improved
add pre-training on COCO images before main training (ablation)Full model · DAVIS 2017 valG: 67.9 -> 81.3Improved
tune proposal generator on small DAVIS instead of COCO-pretrained only (ablation)Full model · DAVIS 2017 valG: 80.1 -> 79.0Hurt
tune proposal generator on YTVOS (ablation)Full model · DAVIS 2017 valG: 80.1 -> 81.3Improved
use DMN with first frame only vs AOA-only base (ablation)Base w/ AOA · DAVIS 2017 valG: 78.9 -> 79.9Improved

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