AI Paper Atlas

Research map / Video / Video segmentation and tracking

Seeing Conversations: Communication Context Identification in Egocentric Video

Tobias Dorszewski, Jens Hjortkjær

CVPR 2026 · Video › Video segmentation and tracking · approach: Layered video decomposition and dynamics

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 temporal LSTM modeling (ablation)CoCoNet · CCI dataset, combined testbalanced accuracy (%): 83 -> 96Improved
add inter-person self-attention (ablation)CoCoNet · CCI dataset, combined testbalanced accuracy (%): 92 -> 96Improved
add both temporal and interaction modules (ablation)CoCoNet · CCI dataset, combined testbalanced accuracy (%): 74 -> 96Improved
combine head direction with face box features (ablation)CoCoNet · CCI dataset, combined testbalanced accuracy (%): 86 (Fbox), 89 (Fdir) -> 94Improved
add detection quality and facial landmarks (ablation)CoCoNet · CCI dataset, combined testbalanced accuracy (%): 94 -> 96Improved
add ResNet features to the structured feature set (ablation)CoCoNet · CCI dataset, combined testbalanced accuracy (%): 96 -> 96No clear effect
train with variable segment lengths instead of fixed 4096 frames (ablation)CoCoNet · CCI dataset, combined testTbAcc (%): 79 -> 87Improved
randomly vary person dimension in training instead of fixed 9 (ablation)CoCoNet · CCI dataset, combined testbalanced accuracy (%): 92 -> 96Improved
train only on small groups (2-3) and test on large groups (ablation)CoCoNet · CCI dataset, group size 6-10balanced accuracy (%): 97 -> 27Hurt
use CoCoNet with self-attention instead of affinity predictorCoCoNet · CCI dataset, combined testbalanced accuracy (%): 90 -> 94 (same features)Improved
longer temporal context at inference for large groups (ablation)CoCoNet · CCI dataset, group size 6-10balanced accuracy (%): 70 -> 97Improved
matched versus unseen individuals at test timeCoCoNet · CCI datasetbalanced accuracy (%): 95 -> 97No clear effect

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