Research map / Graph learning / Knowledge and temporal graphs
Context-Aware Scene Graph Generation With Seq2Seq Transformers
Yichao Lu, Himanshu Rai, Jason Chang, Boris Knyazev, Guangwei Yu, Shashank Shekhar, Graham W. Taylor, Maksims Volkovs
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 sequential decoder conditioned on predicted triplets (teacher forcing) (ablation) | Seq2Seq encoder-only · VG SGDET (graph constraint) | Recall@100: 31.9 -> 33.0 | Improved |
| add sequential decoder conditioned on predicted triplets (teacher forcing) (ablation) | Seq2Seq encoder-only · VRD Relationship Detection (free k) | Recall@100: 31.6 -> 34.4 | Improved |
| add sequential decoder conditioned on predicted triplets (teacher forcing) (ablation) | Seq2Seq encoder-only · VRD Phrase Detection (free k) | Recall@100: 39.9 -> 43.9 | Improved |
| add RL training with mean-recall reward on top of teacher forcing (ablation) | Seq2Seq · VG SGDET (graph constraint) | Recall@100: 33.0 -> 34.4 | Improved |
| add RL training with mean-recall reward on top of teacher forcing (ablation) | Seq2Seq · VG SGDET | mRecall@100: 10.8 -> 12.1 | Improved |
| add RL training with mean-recall reward on top of teacher forcing (ablation) | Seq2Seq · VRD Relationship Detection (free k) | Recall@100: 34.4 -> 35.9 | Improved |
| add RL training with mean-recall reward on top of teacher forcing (ablation) | Seq2Seq · VRD Phrase Detection (free k) | Recall@100: 43.9 -> 46.2 | Improved |
| use RL instead of scheduled sampling (ablation) | Seq2Seq · VG SGDET | mRecall@100: 11.0 -> 12.1 | Improved |
| use RL instead of scheduled sampling (ablation) | Seq2Seq · VG SGDET (graph constraint) | Recall@100: 34.1 -> 34.4 | Improved |
| use scheduled sampling instead of teacher forcing (ablation) | Seq2Seq · VG SGDET (graph constraint) | Recall@100: 33.0 -> 34.1 | Improved |
| increase prediction block size at inference (ablation) | Seq2Seq · VG | accuracy | Hurt |
| vary alpha between recall and mRecall in reward (ablation) | Seq2Seq - RL · VG SGDET | Recall vs mRecall @100 | Mixed |
Closest papers by meaning
- Iterative Scene Graph Generation (NeurIPS 2022 · similarity 0.90)
- IS-GGT: Iterative Scene Graph Generation With Generative Transformers (CVPR 2023 · similarity 0.90)
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 2024 · similarity 0.90)
- Single-Stage Visual Relationship Learning using Conditional Queries (NeurIPS 2022 · similarity 0.90)
- Generate Any Scene: Scene Graph Driven Data Synthesis for Visual Generation Training (ICLR 2026 · similarity 0.89)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.88)
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.
