Research map / Graph learning / Knowledge and temporal graphs
Seq-SG2SL: Inferring Semantic Layout From Scene Graph Through Sequence to Sequence Learning
Boren Li, Boyu Zhuang, Mingyang Li, Jian Gu
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 |
|---|---|---|---|
| use sequential seq-to-seq formulation instead of graph convolution baseline | layout prediction · Visual Genome train | mean-SLEU (IoU 0.5): 0.0217 -> 0.4174 | Improved |
| use sequential seq-to-seq formulation instead of graph convolution baseline | layout prediction · Visual Genome test | mean-SLEU (IoU 0.5): 0.0205 -> 0.0344 | Improved |
| use sequential seq-to-seq formulation instead of graph convolution baseline | layout prediction · Visual Genome test | mean-SLEU (IoU 0.0): 1.0000 -> 0.9998 | No clear effect |
| add relative position encoding for the object box (ablation) | Seq-SG2SL · Visual Genome train | mean-SLEU (IoU 0.5): 0.2698 -> 0.4174 | Improved |
| add relative position encoding for the object box (ablation) | Seq-SG2SL · Visual Genome test | mean-SLEU (IoU 0.5): 0.0334 -> 0.0344 | Improved |
| use Transformer instead of LSTM (ablation) | Seq-SG2SL · Visual Genome train | mean-SLEU (IoU 0.5): 0.3822 -> 0.4174 | Improved |
| use Transformer instead of LSTM (ablation) | Seq-SG2SL · Visual Genome test | mean-SLEU (IoU 0.5): 0.0222 -> 0.0344 | Improved |
| LSTM seq-to-seq vs graph-convolution baseline (ablation) | Seq-SG2SL framework · Visual Genome train | mean-SLEU (IoU 0.5): 0.0217 -> 0.3822 | Improved |
| use beam search top-1 instead of greedy decoding (ablation) | Seq-SG2SL · Visual Genome train | mean-SLEU (IoU 0.5): 0.3958 -> 0.4174 | Improved |
| use beam search top-1 instead of greedy decoding (ablation) | Seq-SG2SL · Visual Genome test | mean-SLEU (IoU 0.5): 0.0335 -> 0.0344 | No clear effect |
| use beam top-2 instead of top-1 candidate (ablation) | Seq-SG2SL · Visual Genome train | mean-SLEU (IoU 0.5): 0.4174 -> 0.3531 | Hurt |
Closest papers by meaning
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 2024 · similarity 0.86)
- Multi-Prototype Space Learning for Commonsense-Based Scene Graph Generation (AAAI 2024 · similarity 0.86)
- SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation (AAAI 2024 · similarity 0.86)
- SGDiff: Scene Graph Guided Diffusion Model for Image Collaborative SegCaptioning (AAAI 2025 · similarity 0.86)
- Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Models (ICML 2026 · similarity 0.86)
- LLM4SGG: Large Language Models for Weakly Supervised Scene Graph Generation (CVPR 2024 · similarity 0.86)
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.
