AI Paper Atlas

Research map / Video / Video-language understanding

Single-Stage Visual Query Localization in Egocentric Videos

Hanwen Jiang, Santhosh Kumar Ramakrishnan, Kristen Grauman

NeurIPS 2023 · Video › Video-language understanding · approach: Egocentric video understanding · 6 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
lower input resolution 448 -> 224 (ablation)VQLoC · Ego4D VQ2D valtAP25: 0.29 -> 0.22Hurt
smaller backbone ViT-b14 -> ViT-s14 (ablation)VQLoC · Ego4D VQ2D valtAP25: 0.29 -> 0.28No clear effect
replace spatial transformer with conv fusion (ablation)VQLoC · Ego4D VQ2D valstAP25: 0.19 -> 0.13Hurt
use 3 spatial TX layers instead of 1 (ablation)VQLoC · Ego4D VQ2D valstAP25: 0.19 -> 0.20No clear effect
use global temporal window instead of local window 5 (ablation)VQLoC · Ego4D VQ2D valtAP25: 0.29 -> 0.08Hurt
local window size 9 instead of 5 (ablation)VQLoC · Ego4D VQ2D valtAP25: 0.29 -> 0.27Hurt
local window size 7 instead of 5 (ablation)VQLoC · Ego4D VQ2D valrec%: 38.58 -> 36.97Hurt
remove anchors, keep BCE loss (ablation)VQLoC · Ego4D VQ2D valtAP25: 0.29 -> 0.17Hurt
anchors with BCE loss instead of focal loss (ablation)VQLoC · Ego4D VQ2D valtAP25: 0.29 -> 0.12Hurt
hard negative mining (BCE+HNM) instead of focal loss (ablation)VQLoC · Ego4D VQ2D valstAP25: 0.19 -> 0.22Improved
single-stage VQLoC vs CocoFormer, medium objectsVQLoC · Ego4D VQ2D val medium objectstAP25: 0.206 -> 0.213Improved
single-stage VQLoC vs CocoFormer, large objectsVQLoC · Ego4D VQ2D val large objectstAP25: 0.338 -> 0.454Improved

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