AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Visual Relationship Detection With Internal and External Linguistic Knowledge Distillation

Ruichi Yu, Ang Li, Vlad I. Morariu, Larry S. Davis

ICCV 2017 · Detection and segmentation › Open-set and few-shot detection · approach: Open-vocabulary 3D and world perception · 329 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 LK distillation (student) to U+W features (ablation)U+W · VRD predicate detection (entire set)R@100/50 (k=1): 37.15 -> 42.98Improved
add LK distillation (student) to U+SF features (ablation)U+SF · VRD predicate detection (entire set)R@100/50 (k=1): 36.33 -> 41.06Improved
add LK distillation (student) to U+W+SF features (ablation)U+W+SF · VRD predicate detection (entire set)R@100/50 (k=1): 41.33 -> 47.50Improved
add LK distillation (student) to U+W+SF features (ablation)U+W+SF · VRD predicate detection (zero-shot)R@100/50 (k=1): 14.13 -> 16.98Improved
add word-embedding semantic features (ablation)U only · VRD predicate detection (entire set)R@100/50 (k=1): 34.82 -> 37.15Improved
add spatial features (ablation)U only · VRD predicate detection (entire set)R@100/50 (k=1): 34.82 -> 36.33Improved
end-to-end fine-tuning of VGG instead of fixed CNN with linear modelBB-Union baseline · VRD predicate detection (entire set)R@100/50 (k=1): 7.11 -> 34.82Improved
combine teacher and student outputs (T+S) instead of teacher only (ablation)U+W+SF+L · VRD predicate detection (entire set)R@100/50 (k=1): 54.13 -> 55.16Improved
use teacher output instead of student on zero-shot relationships (ablation)U+W+SF+L · VRD predicate detection (zero-shot)R@100/50 (k=1): 16.98 -> 8.80Hurt
use linguistic knowledge only, without image (ablation)Baseline: L only · VRD predicate detection (zero-shot)R@100/50 (k=1): 12.75 -> 3.68Hurt
add Wikipedia LK to VRD training-annotation LK (ablation)U+W+SF+L:S · VRD predicate detection, LK source comparison (zero-shot)R@100/50 (k=1): 16.98 -> 19.17Improved
add Visual Genome LK to VRD LK (ablation)U+W+SF+L:S · VRD predicate detection, LK source comparison (zero-shot)R@100/50 (k=1): 16.98 -> 17.21No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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