Research map / Video / Action recognition
Classification Matters: Improving Video Action Detection with Class-Specific Attention
Jinsung Lee, Taeoh Kim, Inwoong Lee, Minho Shim, Dongyoon Wee, Minsu Cho, Suha Kwak
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace vanilla DETR decoder with Localizing Decoder Layer (ablation) | Transformer encoder + decoder · AVA v2.2 (CSN-152) | mAP: 28.6 -> 29.1 | Improved |
| add Classifying Decoder Layer with class queries on top of LDL (ablation) | Transformer encoder + LDL · AVA v2.2 (CSN-152) | mAP: 29.1 -> 31.4 | Improved |
| replace ordinary transformer encoder with 3D deformable multi-scale encoder (ablation) | LDL decoder without CDL · AVA v2.2 (CSN-152) | mAP: 29.1 -> 31.3 | Improved |
| add CDL on top of deformable encoder + LDL (ablation) | full model · AVA v2.2 (CSN-152) | mAP: 31.3 -> 33.5 | Improved |
| attach actor positional queries to class queries (ablation) | CDL · AVA / UCF101-24 | mAP | Improved |
| aggregate multi-scale features conditioned on each actor (ablation) | LDL · AVA / UCF101-24 | mAP | Improved |
| merge actor and context features by simple summation instead of other prior fusions (ablation) | CDL · AVA / UCF101-24 | mAP | Improved |
| give ground-truth class labels vs ground-truth boxes to existing VAD models | TubeR, EVAD, STMixer · AVA | mAP | Improved |
| class-specific attention on JHMDB51-21 (low class diversity) | Ours vs EVAD · JHMDB51-21 | f-mAP: 90.2 -> 86.9 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- STMixer — STMixer: A One-Stage Sparse Action Detector (CVPR 2023) · better on 3 of 3 shared results
- TubeR — TubeR: Tubelet Transformer for Video Action Detection (CVPR 2022) · better on 1 of 1 shared result
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
- Watch Only Once: An End-to-End Video Action Detection Framework (ICCV 2021 · similarity 0.91)
- JoVALE: Detecting Human Actions in Video Using Audiovisual and Language Contexts (AAAI 2025 · similarity 0.88)
- Efficient Video Action Detection with Token Dropout and Context Refinement (ICCV 2023 · similarity 0.88)
- End-To-End Learning of Action Detection From Frame Glimpses in Videos (CVPR 2016 · similarity 0.87)
- Context-Aware RCNN: A Baseline for Action Detection in Videos (ECCV 2020 · similarity 0.87)
- Action Detail Matters: Refining Video Recognition with Local Action Queries (CVPR 2025 · similarity 0.87)
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.
