Research map / Video / Action recognition
Relaxed Transformer Decoders for Direct Action Proposal Generation
Jing Tan, Jiaqi Tang, Limin Wang, Gangshan Wu
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 |
|---|---|---|---|
| add boundary-attentive scaling (alpha=2) to features (ablation) | RTD-Net · THUMOS14 | AR@50: 36.22 -> 41.52 | Improved |
| use MLP encoder instead of Transformer encoder (ablation) | RTD-Net · THUMOS14 | AR@50: 33.69 -> 41.52 | Improved |
| use smaller input receptive field for Transformer encoder (ablation) | RTD-Net with Transformer encoder · THUMOS14 | AR@50: 33.69 -> 36.01 | Improved |
| use smaller input receptive field (64 -> 16) for MLP encoder (ablation) | RTD-Net · THUMOS14 | AR@50: 41.52 -> 39.56 | Hurt |
| add relaxed matcher (ablation) | RTD-Net · ActivityNet-1.3 | AR@100: 71.88 -> 73.21 | Improved |
| add completeness head score to classification score (ablation) | RTD-Net · THUMOS14 | AR@50: 41.08 -> 41.52 | Improved |
| add temporal positional embedding in MLP encoder (ablation) | RTD-Net · THUMOS14 | AR@50: 37.07 -> 41.52 | Improved |
| add positional embedding at decoder attention (ablation) | RTD-Net · THUMOS14 | AR@50: 37.43 -> 41.52 | Improved |
| use pre-enhancement MLP projection instead of post (ablation) | RTD-Net · THUMOS14 | AR@50: 38.81 -> 41.52 | Improved |
| enhance by multiplication instead of concatenation (ablation) | RTD-Net · THUMOS14 | AR@50: 37.92 -> 41.52 | Improved |
| increase decoder layers from 1 to 2 (ablation) | RTD-Net · THUMOS14 | AR@50: 32.76 -> 38.92 | Improved |
| increase decoder layers from 6 to 9 (ablation) | RTD-Net · THUMOS14 | AR@50: 41.52 -> 38.36 | Hurt |
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.
- G-TAD — G-TAD: Sub-Graph Localization for Temporal Action Detection (CVPR 2020) · 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
- Proposal-Free Temporal Action Detection via Global Segmentation Mask Learning (ECCV 2022 · similarity 0.89)
- Temporal Context Aggregation Network for Temporal Action Proposal Refinement (CVPR 2021 · similarity 0.89)
- DiffTAD: Temporal Action Detection with Proposal Denoising Diffusion (ICCV 2023 · similarity 0.89)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.88)
- HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding (NeurIPS 2026 · similarity 0.87)
- Dual DETRs for Multi-Label Temporal Action Detection (CVPR 2024 · 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.
