Research map / Video / Video segmentation and tracking
Dynamic Group Detection using VLM-augmented Temporal Groupness Graph
Kaname Yokoyama, Chihiro Nakatani, Norimichi Ukita
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 two-circle visual prompt instead of no prompt (ablation) | GA-CLIP · JRDB static | F1: 0.735 -> 0.790 | Improved |
| use two-circle visual prompt instead of mask prompt (ablation) | GA-CLIP · JRDB static | F1: 0.728 -> 0.790 | Improved |
| use two-circle visual prompt instead of a single circle (ablation) | GA-CLIP · Cafe static | F1: 0.722 -> 0.819 | Improved |
| add occlusion class to CLIP finetuning (ablation) | GA-CLIP · JRDB dynamic | F1: 0.686 -> 0.769 | Improved |
| add occlusion class to CLIP finetuning (ablation) | GA-CLIP · JRDB static | F1: 0.760 -> 0.790 | Improved |
| use GA-CLIP image encoder instead of DINOv2 (ablation) | group detection · JRDB static | F1: 0.736 -> 0.790 | Improved |
| use GA-CLIP image encoder instead of ResNet50 (ablation) | group detection · Cafe dynamic | F1: 0.634 -> 0.776 | Improved |
| add appearance features to trajectory features (ablation) | group detection · JRDB static | F1: 0.697 -> 0.790 | Improved |
| add trajectory features to appearance features (ablation) | group detection · Cafe dynamic | F1: 0.711 -> 0.776 | Improved |
| use Louvain clustering instead of label propagation | temporal groupness graph · JRDB dynamic | F1: 0.601 -> 0.769 | Improved |
| use Louvain clustering instead of label propagation for S3R2 | S3R2 · JRDB dynamic | F1: 0.584 -> 0.502 | Hurt |
| use tracker-estimated boxes instead of annotated boxes | Ours + Louvain · JRDB dynamic | F1: 0.769 -> 0.774 | Mixed |
Closest papers by meaning
- A Multiview-Based Parameter Free Framework for Group Detection (AAAI 2017 · similarity 0.86)
- Detecting Coherent Groups in Crowd Scenes by Multiview Clustering (TPAMI 2018 · similarity 0.85)
- Detector-Free Weakly Supervised Group Activity Recognition (CVPR 2022 · similarity 0.85)
- Learning Multi-Attention Context Graph for Group-Based Re-Identification (TPAMI 2020 · similarity 0.85)
- MINGLE: VLMs for Semantically Complex Region Detection in Urban Scenes (AAAI 2026 · similarity 0.85)
- Perceptual Group Tokenizer: Building Perception with Iterative Grouping (ICLR 2024 · similarity 0.82)
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.
