Research map / Video / Action recognition
Self-Feedback DETR for Temporal Action Detection
Jihwan Kim, Miso Lee, Jae-Pil Heo
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 self-feedback losses for both encoder and decoder self-attention (ablation) | DETR for TAD · THUMOS14 | mAP avg: 50.3 -> 56.7 | Improved |
| add decoder self-feedback only (ablation) | DETR for TAD · THUMOS14 | mAP avg: 50.3 -> 54.4 | Improved |
| add encoder self-feedback only (ablation) | DETR for TAD · THUMOS14 | mAP avg: 50.3 -> 53.7 | Improved |
| aggregate encoder self-attention by matmul instead of last-layer-only feedback (ablation) | Self-DETR encoder feedback · THUMOS14 | mAP avg: 54.3 -> 56.7 | Improved |
| give per-layer decoder guidance instead of last or averaged guidance (ablation) | Self-DETR decoder feedback · THUMOS14 | mAP avg: 51.1 -> 56.7 | Improved |
| use identity or relative-attention guidance instead of cross-attention guidance (ablation) | Self-DETR · THUMOS14 | mAP avg: 56.7 -> 49.9 (identity), 50.3 (relative) | Hurt |
| maximize attention diversity directly instead of cross-attention guidance (ablation) | Self-DETR · THUMOS14 | mAP avg: 50.3 -> 37.1 | Hurt |
| lower decoder loss weight from 5 to 1 (ablation) | Self-DETR · THUMOS14 | mAP avg: 56.7 -> 54.2 | Hurt |
| remove self-attention in encoder and decoder from baseline DETR (ablation) | DETR for TAD baseline · THUMOS14 | mAP avg: 50.3 -> 48.3 | No clear effect |
| add self-feedback to DINO for TAD | DINO · THUMOS14 | mAP avg: 50.9 -> 56.7 | Improved |
| reduce decoder layers from 4 to 1 (ablation) | Self-DETR · THUMOS14 | mAP avg: 56.7 -> 46.1 | Hurt |
| Self-DETR vs ReAct at high IoU | DETR-based TAD · THUMOS14 | mAP@0.7: 35.6 -> 31.8 | Mixed |
Closest papers by meaning
- Prediction-Feedback DETR for Temporal Action Detection (AAAI 2025 · similarity 0.94)
- Dual DETRs for Multi-Label Temporal Action Detection (CVPR 2024 · similarity 0.90)
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.88)
- TE-TAD: Towards Full End-to-End Temporal Action Detection via Time-Aligned Coordinate Expression (CVPR 2024 · similarity 0.86)
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (CVPR 2024 · similarity 0.85)
- DiGIT: Multi-Dilated Gated Encoder and Central-Adjacent Region Integrated Decoder for Temporal Action Detection Transformer (CVPR 2025 · similarity 0.85)
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.
