AI Paper Atlas

Research map / Video / Video segmentation and tracking

You Can Ground Earlier Than See: An Effective and Efficient Pipeline for Temporal Sentence Grounding in Compressed Videos

Xiang Fang, Daizong Liu, Pan Zhou, Guoshun Nan

CVPR 2023 · Video › Video segmentation and tracking · approach: Self-supervised video representation learning · 49 citations (OpenAlex)

Read the paper

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 pseudo feature generation to baseline (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.3: 50.59 -> 60.25Improved
add three-branch spatial-temporal attention to baseline (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.3: 50.59 -> 62.79Improved
add adaptive motion-appearance fusion to baseline (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.3: 50.59 -> 63.74Improved
add adaptive fusion on top of PFG and TTA (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.3: 64.19 -> 66.87Improved
add motion vector and residual features to I-frame only (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.3: 64.27 -> 66.87Improved
use noun-formed query in spatial attention (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.3: 64.26 -> 66.87Improved
add temporal attention to spatial attention (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.5: 43.20 -> 48.38Improved
add pseudo appearance feature to pseudo motion feature (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.5: 47.51 -> 48.38Improved
increase temporal attention window K from 7 to 8 (ablation)TCSF · ActivityNet CaptionsR@1 IoU=0.3: 66.87 -> 65.92Hurt
use compressed-domain TCSF instead of fully decoded fully-supervised MMNTCSF · Charades-STAR@1 IoU=0.7: 27.28 -> 37.20Improved
use compressed-domain TCSF instead of fully decoded MMNTCSF · ActivityNet CaptionsR@1 IoU=0.5: 48.59 vs 48.38No clear effect
use compressed-domain TCSF instead of decoded-frame TAG for inference timeTCSF · ActivityNet Captionstotal time of 100 videos (s): 243.36 -> 46.16Improved

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