Research map / Video / Video-language understanding
Revisiting Weakly-Supervised Video Scene Graph Generation via Pair Affinity Learning
MINSEOK KANG, Minhyeok Lee, Minjung Kim, Jungho Lee, Donghyeong Kim, Sungmin Woo, Inseok Jeon, Sangyoun Lee
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 pair affinity learning and scoring (PALS) (ablation) | PLA + STTran · AG SGDet (With Constraint) | R@10: 15.39 -> 19.79 | Improved |
| add pair affinity modulation (PAM) to PALS (ablation) | PLA + STTran · AG SGDet (With Constraint) | R@10: 19.79 -> 20.22 | Improved |
| add relation-aware matching (RAM) alone (ablation) | PLA + STTran · AG SGDet (With Constraint) | R@10: 15.39 -> 15.90 | Improved |
| add RAM on top of PALS (ablation) | PLA + STTran · AG SGDet (With Constraint) | R@10: 19.79 -> 22.14 | Improved |
| add PAM on top of RAM + PALS (ablation) | PLA + STTran · AG SGDet (With Constraint) | R@10: 22.14 -> 22.24 | Improved |
| add PALS to the No Constraint setting (ablation) | PLA + STTran · AG SGDet (No Constraint) | R@10: 15.83 -> 21.03 | Improved |
| use pair affinity score in inference-time ranking (ablation) | PLA + Ours (STTran) · AG SGDet | R@10: +5.16 in (W)R@10 and +6.95 in (N)R@10 | Improved |
| replace class-level matching with RAM (GroundingDINO) (ablation) | PLA pseudo-labels · AG pseudo-label quality | precision: 0.4016 -> 0.7255 | Improved |
| replace class-level matching with RAM (GroundingDINO) (ablation) | PLA pseudo-labels · AG pseudo-label quality | recall: 0.4423 -> 0.3995 | Hurt |
| use GLIP instead of GroundingDINO in RAM (ablation) | PLA + Ours (STTran) · AG SGDet (With Constraint) | R@50: 28.00 -> 26.65 | Hurt |
| lower RAM reliability threshold fallback: tau_r 0.3 vs 1.0 (class-level matching) (ablation) | PLA + Ours (STTran) · AG SGDet (With Constraint) | R@10: 20.25 -> 22.24 | Improved |
| raise grounding score threshold tau_gs from 0.2 to 0.9 (ablation) | PLA + Ours (STTran) · AG SGDet (With Constraint) | R@10: 22.24 -> 19.82 | 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.
- TRKT — TRKT: Weakly Supervised Dynamic Scene Graph Generation with Temporal-enhanced Relation-aware Knowledge Transferring (ICCV 2025) · better on 2 of 2 shared results
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
- TRKT: Weakly Supervised Dynamic Scene Graph Generation with Temporal-enhanced Relation-aware Knowledge Transferring (ICCV 2025 · similarity 0.91)
- A Simple Baseline for Weakly-Supervised Scene Graph Generation (ICCV 2021 · similarity 0.91)
- A Closer Look at Dynamic Scene Graph Generation in the Era of Multimodal Large Language Models (NeurIPS 2026 · similarity 0.88)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.88)
- DiffVsgg: Diffusion-Driven Online Video Scene Graph Generation (CVPR 2025 · similarity 0.87)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (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.
