Research map / Vision-language and multimodal / Cross-modal learning
Diff-MM: Exploring Pre-Trained Text-to-Image Generation Models for Unified Multi-Modal Object Tracking
Shiyu Xuan, Zechao Li, Jinhui Tang, Maocheng Zhao
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 parallel feature extraction (PFE) instead of Diff-Tracker prompt tokens (ablation) | Diff-MM · LaSOT | AUC: 58.7 -> 71.5 | Improved |
| use PFE instead of cross-attention with MAE template encoder (ablation) | Diff-MM · LaSOT | AUC: 65.7 -> 71.5 | Improved |
| use PFE instead of cross-attention with CLIP ViT template encoder (ablation) | Diff-MM · LaSOT | AUC: 64.5 -> 71.5 | Improved |
| use PFE instead of Diff-Tracker (ablation) | Diff-MM · TrackingNet | AUC: 75.6 -> 84.7 | Improved |
| add auxiliary depth modality via multi-modal sub-module tuning (ablation) | Diff-MM · DepthTrack | F-score: 58.5 -> 65.3 | Improved |
| use sub-module instead of early fusion (ablation) | Diff-MM · DepthTrack | F-score: 60.9 -> 65.3 | Improved |
| reuse VAE encoder instead of T2I-Adapter extra encoder (ablation) | Diff-MM · LasHeR | PR: 63.7 -> 71.3 | Improved |
| freeze UNet in second stage instead of tuning it (ablation) | Diff-MM · DepthTrack | F-score: 62.3 -> 65.3 | Improved |
| add zero-init convolution to lateral connections (ablation) | Diff-MM · DepthTrack | F-score: 62.6 -> 65.3 | Improved |
| add TNL2K to first-stage training data (ablation) | Diff-MM · DepthTrack | F-score: 65.5 -> 65.3 | No clear effect |
| use SD v3-5 (MM-DiT) as base model instead of SD v1-5 | Diff-MM 256 · TNL2K | AUC: 65.2 -> 67.5 | Improved |
| use single parameter set instead of modality-specific parameters | Diff-MM 256 · DepthTrack | F-score: 65.3 -> 65.9 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SDSTrack — SDSTrack: Self-Distillation Symmetric Adapter Learning for Multi-Modal Visual Object Tracking (CVPR 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Collaborating Vision, Depth, and Thermal Signals for Multi-Modal Tracking: Dataset and Algorithm (NeurIPS 2025 · similarity 0.91)
- Single-Model and Any-Modality for Video Object Tracking (CVPR 2024 · similarity 0.90)
- Bi-directional Adapter for Multimodal Tracking (AAAI 2024 · similarity 0.89)
- SDSTrack: Self-Distillation Symmetric Adapter Learning for Multi-Modal Visual Object Tracking (CVPR 2024 · similarity 0.88)
- HFF-Tracker: A Hierarchical Fine-grained Fusion Tracker for Referring Multi-Object Tracking (AAAI 2025 · similarity 0.87)
- Generative-Based Fusion Mechanism for Multi-Modal Tracking (AAAI 2024 · similarity 0.87)
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.
