AI Paper Atlas

Research map / Video / Video segmentation and tracking

SalSAC: A Video Saliency Prediction Model with Shuffled Attentions and Correlation-Based ConvLSTM

Xinyi Wu, Zhenyao Wu, Jinglin Zhang, Lili Ju, Song Wang

AAAI 2020 · Video › Video segmentation and tracking · approach: Stochastic video prediction · 71 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 attention (ablation)SalSAC · DHF1K valNSS: 2.580 -> 2.624Improved
add random shuffling of multi-level attention maps (vs unshuffled multi-scale) (ablation)SalSAC · DHF1K valNSS: 2.613 -> 2.624Improved
add random shuffling of multi-level attention maps (ablation)SalSAC · DHF1K vals-AUC: 0.700 -> 0.729Improved
use multi-scale instead of single-scale attention (ablation)SalSAC · DHF1K valNSS: 2.600 -> 2.613Improved
add ConvLSTM (ablation)SalSAC · DHF1K valNSS: 2.461 -> 2.624Improved
correlation-based ConvLSTM instead of traditional ConvLSTM (ablation)SalSAC · DHF1K valNSS: 2.547 -> 2.624Improved
balance both input and output of ConvLSTM (vs v1 output only) (ablation)SalSAC · DHF1K vals-AUC: 0.709 -> 0.729Improved
balance both input and output (vs v2 input only) (ablation)SalSAC · DHF1K vals-AUC: 0.696 -> 0.729Improved
add SALICON image pre-training (ablation)SalSAC · DHF1K valAUC-J: 0.890 -> 0.898Improved
SalSAC vs STRA-NetSalSAC · Hollywood-2NSS: 3.478 -> 3.356Mixed

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