Research map / Architectures and efficiency / Attention and transformers
Generalized Relation Modeling for Transformer Tracking
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the two-stream relation pipeline with GRM adaptive token division (ablation) | Transformer tracking · GOT-10k* | AO: 64.7 -> 73.4 | Helped |
| Allow early template-search interaction for all search tokens (one-stream instead of two-stream) (ablation) | Transformer tracking · GOT-10k* | AO: 64.7 -> 71.0 | Helped |
| Add an additional shared search-token category E_A to binary token division (ablation) | GRM · GOT-10k* | AO: 71.5 -> 71.9 | Helped |
| Divide search tokens into E_S and E_A, preserving interaction among all search tokens while selecting template-interacting tokens (ablation) | GRM · GOT-10k* | AO: 71.9 -> 73.4 | Helped |
| Apply token division in the first encoder layer (ablation) | GRM · GOT-10k* | AO: 73.4 -> 71.2 | Hurt |
| Restrict token division to layers 5–12 rather than layers 2–12 (ablation) | GRM · GOT-10k* | AO: 73.4 -> 72.3 | Hurt |
| Use maximum pooling rather than average pooling for the target-aware representation (ablation) | GRM · GOT-10k* | AO: 71.8 -> 73.4 | Helped |
| Use attention masking to parallelize the attention operations (ablation) | GRM inference · Tracking inference | fps: 33 -> 45 | Helped |
| Use GRM instead of OSTrack | Transformer tracking · GOT-10k* | AO: 71.0 -> 73.4 | Helped |
| Use GRM instead of OSTrack | Transformer tracking · TrackingNet | AUC: 83.1 -> 84.0 | Helped |
| Use GRM instead of OSTrack | Transformer tracking · LaSOT | AUC: 69.1 -> 69.9 | Helped |
| Use GRM instead of MixFormer | Transformer tracking · AVisT | AUC: 53.7 -> 54.5 | Helped |
Closest papers by meaning
- Target Focused Shallow Transformer Framework for Efficient Visual Tracking (AAAI 2024 · similarity 0.86)
- UTPTrack: Towards Simple and Unified Token Pruning for Visual Tracking (CVPR 2026 · similarity 0.85)
- Is Multiple Object Tracking a Matter of Specialization? (NeurIPS 2024 · similarity 0.84)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.83)
- Sequential Fusion Based Multi-Granularity Consistency for Space-Time Transformer Tracking (AAAI 2024 · similarity 0.83)
- Cross-modulated Attention Transformer for RGBT Tracking (AAAI 2025 · 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.