AI Paper Atlas

Research map / Video / Action recognition

BMN: Boundary-Matching Network for Temporal Action Proposal Generation

Tianwei Lin, Xiao Liu, Xin Li, Errui Ding, Shilei Wen

ICCV 2019 · Video › Action recognition · approach: Fully supervised temporal action detection · 728 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 PEM trained separately to TEM (ablation)BMN · ActivityNet-1.3 valAUC: 65.17 -> 66.43Improved
add PEM trained separately to TEM (ablation)BSN · ActivityNet-1.3 valAUC: 64.80 -> 66.17Improved
train TEM and PEM jointly end-to-end (ablation)BMN · ActivityNet-1.3 valAR@100: 74.36 -> 75.01Improved
train TEM and PEM jointly end-to-end (ablation)BMN · ActivityNet-1.3 valAUC: 66.43 -> 67.10Improved
train TEM and PEM jointly end-to-end (ablation)BMN · ActivityNet-1.3 valtotal inference time (s): 0.069 -> 0.052Improved
replace BSN proposal-feature scoring with boundary-matching confidence mapBSN · ActivityNet-1.3 valtotal inference time (s): 0.629 -> 0.069Improved
replace BSN with BMN (TEM only) (ablation)BSN TEM · ActivityNet-1.3 valAR@100: 73.57 -> 73.72No clear effect
replace BSN with BMN (end-to-end)BSN · ActivityNet-1.3 valAUC: 66.17 -> 67.10Improved
use Soft-NMS instead of Greedy-NMS (ablation)BMN · THUMOS-14 (C3D features)AR@50: 29.04 -> 32.73Improved
use Soft-NMS instead of Greedy-NMS (ablation)BMN · THUMOS-14 (C3D features)AR@1000: 60.96 -> 60.44Hurt
train on seen+unseen categories instead of seen only (ablation)BMN · ActivityNet-1.3 unseen subsetAUC: 64.47 -> 65.05No clear effect
use C3D to two-stream featuresBMN+SNMS · THUMOS-14AR@100: 40.68 -> 47.72Improved

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