AI Paper Atlas

Research map / Video / Action recognition

Weakly-Supervised Action Localization, and Action Recognition Using Global–Local Attention of 3D CNN

Novanto Yudistira, Muthu Subash Kavitha, Takio Kurita

IJCV 2022 · Video › Action recognition · approach: Weakly supervised temporal action localization · 15 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
aggregate 3DCAM over all layers instead of GradCAM on last layer3D ResNeXt-101 · JHMDB pointing gameaccuracy: 0.988 -> 0.991No clear effect
use Sum 3DCAM instead of GradCAM3D ResNeXt-101 · JHMDB localization IoU 0.5 and 0.6localization accuracy: 58.75 % and 284.61 % improvement at IoU 0.5 and 0.6Improved
use Sum 3DCAM instead of GradCAM for joint classification and localization3D ResNeXt-101 · JHMDB Acc2Acc2: 46.15 % and 90.0 % improvement at IoU 0.5 and 0.6Improved
sum aggregation instead of average pooling of layer 3DCAMs (ablation)3DCAM · JHMDB localizationlocalization accuracy: 0.089 -> 0.127 at IoU 0.5Improved
use only single layers (first or fourth) rather than all layers (ablation)3DCAM on JHMDB · JHMDB pointing gameaccuracy: 0.991 -> 0.821Hurt
aggregate conv and layer 1 instead of conv layer alone (ablation)3DCAM · UCF24MaxBoxAcc average: 0.159 -> 0.395Improved
aggregate layers conv and 1 instead of a single layer 1 (ablation)3DCAM · UCF24MaxBoxAcc average: 0.366 -> 0.395Improved
add layers 2 and 3 to the conv+layer1 aggregation (ablation)Sum 3DCAM · UCF24MaxBoxAcc average: 0.395 -> 0.37Hurt
use 3DCAM instead of GradCAM with same layer compositionSum aggregation (conv and layer 1) · UCF24MaxBoxAcc average: 0.361 -> 0.395 (2.5% difference stated)Improved
aggregate 3DCAM layers 2,3,4 instead of GradCAM layer 4 onlyResNet50 · ILSVRC localizationMaxBoxAcc average: 0.57 -> 0.598Improved
3DCAM sum instead of Sum GradCAM at IoU 0.5ResNet50 · ILSVRC localizationMaxBoxAcc IoU 0.5: 0.622 -> 0.622No clear effect
add layer-wise attention gates to 3D ResNeXt-1013D ResNeXt-101 · HMDB51video-wise accuracy: 0.891 -> 0.919Improved

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