Research map / Graph learning / Knowledge and temporal graphs
DSFlash: Comprehensive Panoptic Scene Graph Generation in Realtime
Julian Lorenz, Vladyslav Kovganko, Elias Kohout, Mrunmai Phatak, Daniel Kienzle, Rainer Lienhart
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 |
|---|---|---|---|
| replace two backbones (MaskDINO + ResNet) with single frozen EoMT backbone (ablation) | DSFormer · PSG SGDet | latency: 445 ms -> 41 ms | Improved |
| replace two backbones with single EoMT backbone (ablation) | DSFormer · PSG SGDet | mR@50: 30.7 -> 25.0 | Hurt |
| add naive bidirectional prediction (one pass, both directions) (ablation) | DSFlash · PSG SGDet | mR@50: 25.0 -> 23.8 | Hurt |
| add gate mechanism and feature consistency loss to bidirectional prediction (ablation) | DSFlash · PSG SGDet | mR@50: 23.8 -> 28.8 | Improved |
| skip upsampling of segmentation mask logits (ablation) | DSFlash · PSG SGDet | latency: 29 ms -> 23 ms | Improved |
| skip upsampling of segmentation mask logits (ablation) | DSFlash · PSG SGDet | mR@50: 28.8 -> 28.5 | Hurt |
| use efficient mask embedding implementation (ablation) | DSFlash · PSG SGDet | mR@50: 25.0 -> 25.0 | No clear effect |
| use larger EoMT-L backbone instead of EoMT-S (ablation) | DSFlash · PSG SGDet | mR@50: 25.1 -> 30.9 | Improved |
| add dynamic patch pruning (no retraining) (ablation) | DSFlash · PSG SGDet | mR@50: slight decrease of 2.13 | Hurt |
| add patch pruning (ablation) | DSFlash · PSG, GTX 1080 | latency: 230 ms -> 205 ms | Improved |
| add patch pruning on high-end GPUs (H100, RTX 3090) at batch size 1 (ablation) | DSFlash · PSG | latency | No clear effect |
| combine token merging with patch pruning (ablation) | DSFlash · PSG, GTX 1080 | latency: 230 ms -> 173 ms | Improved |
Closest papers by meaning
- Towards Unbiased and Robust Spatio-Temporal Scene Graph Generation and Anticipation (CVPR 2025 · similarity 0.87)
- HiKER-SGG: Hierarchical Knowledge Enhanced Robust Scene Graph Generation (CVPR 2024 · similarity 0.87)
- HyperGLM: HyperGraph for Video Scene Graph Generation and Anticipation (CVPR 2025 · similarity 0.86)
- DiffVsgg: Diffusion-Driven Online Video Scene Graph Generation (CVPR 2025 · similarity 0.86)
- A Closer Look at Dynamic Scene Graph Generation in the Era of Multimodal Large Language Models (NeurIPS 2026 · similarity 0.86)
- Panoptic Scene Graph Generation with Semantics-Prototype Learning (AAAI 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.
