AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Probabilistic Attention for Interactive Segmentation

Prasad Gabbur, Manjot Bilkhu, Javier Movellan

NeurIPS 2021 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 1 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
Adapt keys online from observed queriesProbabilistic BoTNet50 · GrabCut and Berkeley, low-feedback regimemIoU: ~10% absolute improvement without corrective clicksImproved
Use key adaptation with lower prior precision, including maximum-likelihood adaptation (ablation)Probabilistic BoTNet50-FactoredPE · GrabCut and BerkeleymIoUImproved
Use full rather than factored relative position encoding (ablation)Probabilistic BoTNet50 · GrabCut and BerkeleymIoUImproved
Replace standard backbone attention with probabilistic attentionBoTNet50 · GrabCut and Berkeley, especially with fewer clicksmIoUImproved
Propagate fixed annotator-provided values onlineCorrective self-attention classification head · GrabCut and Berkeley, high-feedback regimemIoUImproved
Use key adaptation and value propagation togetherProbabilistic BoTNet50 with corrective self-attention head · GrabCut and Berkeley, low- and high-feedback regimesmIoUImproved
Use value propagation in both axial-attention height and width blocks instead of only the width block (ablation)Corrective self-attention classification head · Interactive segmentationmIoUHurt
Use value propagation for multiple iterations (1 or 5 iterations were tested) (ablation)Corrective self-attention classification head · GrabCut and BerkeleymIoUImproved

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