Research map / Graph learning / Knowledge and temporal graphs
TextPSG: Panoptic Scene Graph Generation from Textual Descriptions
Chengyang Zhao, Yikang Shen, Zhenfang Chen, Mingyu Ding, Chuang Gan
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 TextPSG instead of SGGNLS-o (which uses a pre-trained detector) | Caption-to-PSG · PhrDet PSG | N5R100: 7.93 -> 14.37 | Improved |
| add graph-cut merging on stage-1 segments (ablation) | Segment merger · PhrDet PSG | N5R100: 11.39 -> 14.37 | Improved |
| apply graph cut on stage-2 (8) segments instead of stage-1 (ablation) | Segment merger · PhrDet PSG | N5R100: 11.03 -> 8.45 | Hurt |
| generation-based label prediction instead of classification with WordNet (RNN) (ablation) | Label generator · PhrDet PSG | N5R100: 9.36 -> 10.44 | Improved |
| use pre-trained BLIP decoder instead of RNN (with PET) (ablation) | Label generator · PhrDet PSG | N5R100: 10.44 -> 14.28 | Improved |
| add prompt-embedding-based technique (PET) to BLIP decoder (ablation) | Label generator · PhrDet PSG | N5R100: 2.58 -> 14.28 | Improved |
| evaluate on OOD triplets instead of ID triplets | TextPSG vs SGGNLS-c · PSG OOD split PhrDet | N5R100: TextPSG 14.82 -> 11.69; SGGNLS-c 18.48 -> 0 | Improved |
| replace GroupViT multi-label contrastive loss with entity grounder and segment merger | GroupViT · COCO TSSS | mIoU: 24.72 -> 26.87 | Improved |
Closest papers by meaning
- Scene Graph Generation with Role-Playing Large Language Models (NeurIPS 2024 · similarity 0.89)
- Panoptic Scene Graph Generation with Semantics-Prototype Learning (AAAI 2024 · similarity 0.89)
- Universal Scene Graph Generation (CVPR 2025 · similarity 0.89)
- The Scene Language: Representing Scenes with Programs, Words, and Embeddings (CVPR 2025 · similarity 0.88)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.88)
- LLM4SGG: Large Language Models for Weakly Supervised 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.
