Research map / Video / Video segmentation and tracking
Synthetic Visual Genome 2: Extracting Large-scale Spatio-Temporal Scene Graphs from Videos
Ziqi Gao, Jieyu Zhang, Wisdom Ikezogwo, Jae Sung Park, Tario You, Daniel Ogbu, Chenhao Zheng, Weikai Huang, Yinuo Yang, Winson Han, Quan Kong, Rajat Saini and 1 more
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 trajectory-aligned dual-resampler design with Qwen2.5-VL-3B instead of plain fine-tuning on bbox trajectories | Qwen2.5-VL-3B · SVG2_test (lenient) | object accuracy: 46.1 -> 79.0 | Improved |
| use TraSeR instead of GPT-5 | video scene graph generation · SVG2_test (lenient) | object accuracy: 65.5 -> 79.0 | Improved |
| use TraSeR instead of GPT-5 for relations | video scene graph generation · PVSG (lenient, IoU 0.5) | relation recall: 18.3 -> 16.9 | Hurt |
| add temporal-window resampler on top of object-trajectory resampler (ablation) | TraSeR · PVSG* (VidOR subset) | relation recall: 2.68 -> 11.38 | Improved |
| add object-trajectory resampler on top of temporal-window resampler (ablation) | TraSeR · PVSG* (VidOR subset) | object accuracy: 77.84 -> 78.21 | No clear effect |
| add object-trajectory resampler alone (no window resampler) (ablation) | TraSeR · PVSG* (VidOR subset) | relation recall: 7.15 -> 2.68 | Hurt |
| use segmentation instead of bounding-box token arrangement (ablation) | TraSeR · PVSG* (VidOR subset) | relation recall: 9.39 -> 11.38 | Improved |
| train on SVG2 (SA-V) in addition to academic datasets (ablation) | TraSeR · SVG2_test | object accuracy: 25.68 -> 78.14 | Improved |
| add PVD subset to training data (ablation) | TraSeR · SVG2_test | attribute recall: 22.72 -> 27.10 | Improved |
| train on academic only vs full data for VidOR relations (ablation) | TraSeR · VidOR | relation recall: 28.66 -> 25.01 | Hurt |
| add TraSeR scene graphs to video input for GPT-4.1 | GPT-4.1 VQA · Perception-Test | accuracy: 66.8 -> 71.4 | Improved |
| add TraSeR scene graphs to video input for GPT-4.1 | GPT-4.1 VQA · AGQA 2.0 | accuracy: 25.9 -> 26.3 | Improved |
Closest papers by meaning
- TD²-Net: Toward Denoising and Debiasing for Video Scene Graph Generation (AAAI 2024 · similarity 0.89)
- DiffVsgg: Diffusion-Driven Online Video Scene Graph Generation (CVPR 2025 · similarity 0.89)
- Towards Unbiased and Robust Spatio-Temporal Scene Graph Generation and Anticipation (CVPR 2025 · similarity 0.88)
- HiKER-SGG: Hierarchical Knowledge Enhanced Robust Scene Graph Generation (CVPR 2024 · similarity 0.88)
- A Closer Look at Dynamic Scene Graph Generation in the Era of Multimodal Large Language Models (NeurIPS 2026 · similarity 0.88)
- OED: Towards One-stage End-to-End Dynamic 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.
