Papers › TemporalMaxer: Maximize Temporal Context with only Max Pooling for Temporal Action Localization

TemporalMaxer: Maximize Temporal Context with only Max Pooling for Temporal Action Localization

16 Mar 2023arXiv:2303.09055archive 2025-07-28

Tuan N. Tang, Kwonyoung Kim, Kwanghoon Sohn

Temporal Action Localization (TAL) is a challenging task in video understanding that aims to identify and localize actions within a video sequence. Recent studies have emphasized the importance of applying long-term temporal context modeling (TCM) blocks to the extracted video clip features such as employing complex self-attention mechanisms. In this paper, we present the simplest method ever to address this task and argue that the extracted video clip features are already informative to achieve outstanding performance without sophisticated architectures. To this end, we introduce TemporalMaxer, which minimizes long-term temporal context modeling while maximizing information from the extracted video clip features with a basic, parameter-free, and local region operating max-pooling block. Picking out only the most critical information for adjacent and local clip embeddings, this block results in a more efficient TAL model. We demonstrate that TemporalMaxer outperforms other state-of-the-art methods that utilize long-term TCM such as self-attention on various TAL datasets while requiring significantly fewer parameters and computational resources. The code for our approach is publicly available at https://github.com/TuanTNG/TemporalMaxer

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

tuantng/temporalmaxer officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Action LocalizationTemporal Action LocalizationVideo Understanding

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Temporal Action Localization EPIC-KITCHENS-100 TemporalMaxer (verb) Avg mAP (0.1-0.5) 24.5 #3 of 6 Archive leaderboard report
Temporal Action Localization EPIC-KITCHENS-100 TemporalMaxer (verb) mAP IOU@0.1 27.8 #3 of 6 Archive leaderboard report
Temporal Action Localization EPIC-KITCHENS-100 TemporalMaxer (verb) mAP IOU@0.2 26.6 #3 of 6 Archive leaderboard report
Temporal Action Localization EPIC-KITCHENS-100 TemporalMaxer (verb) mAP IOU@0.3 25.3 #3 of 6 Archive leaderboard report
Temporal Action Localization EPIC-KITCHENS-100 TemporalMaxer (verb) mAP IOU@0.4 23.1 #3 of 6 Archive leaderboard report
Temporal Action Localization EPIC-KITCHENS-100 TemporalMaxer (verb) mAP IOU@0.5 19.9 #3 of 6 Archive leaderboard report
Temporal Action Localization MUSES TemporalMaxer mAP 27.2 #1 of 2 Archive leaderboard report
Temporal Action Localization MUSES TemporalMaxer mAP@0.3 36.7 #1 of 2 Archive leaderboard report
Temporal Action Localization MUSES TemporalMaxer mAP@0.4 33.2 #1 of 2 Archive leaderboard report
Temporal Action Localization MUSES TemporalMaxer mAP@0.5 27.8 #1 of 2 Archive leaderboard report
Temporal Action Localization MUSES TemporalMaxer mAP@0.6 21.9 #1 of 2 Archive leaderboard report
Temporal Action Localization MUSES TemporalMaxer mAP@0.7 16.2 #1 of 2 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer Average mAP 29.9 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.1 49.1 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.2 47.5 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.3 44.3 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.4 39.4 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.5 33.4 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.6 26.5 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.7 17.4 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.8 9.1 #4 of 8 Archive leaderboard report
Temporal Action Localization MultiTHUMOS TemporalMaxer mAP IOU@0.9 2.24 #4 of 8 Archive leaderboard report
Temporal Action Localization THUMOS’14 TemporalMaxer (I3D features) Avg mAP (0.3:0.7) 67.7 #11 of 42 Archive leaderboard report
Temporal Action Localization THUMOS’14 TemporalMaxer (I3D features) mAP IOU@0.3 82.8 #11 of 42 Archive leaderboard report
Temporal Action Localization THUMOS’14 TemporalMaxer (I3D features) mAP IOU@0.4 78.9 #11 of 42 Archive leaderboard report
Temporal Action Localization THUMOS’14 TemporalMaxer (I3D features) mAP IOU@0.5 71.8 #11 of 42 Archive leaderboard report
Temporal Action Localization THUMOS’14 TemporalMaxer (I3D features) mAP IOU@0.6 60.5 #11 of 42 Archive leaderboard report
Temporal Action Localization THUMOS’14 TemporalMaxer (I3D features) mAP IOU@0.7 44.7 #11 of 42 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

CLIP

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections