Research map / Detection and segmentation / Tracking
Representation Learning for Visual Object Tracking by Masked Appearance Transfer
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 |
|---|---|---|---|
| Train jointly encoded template and search features with masked appearance transfer rather than reconstructing the original inputs (ablation) | ViT-B/16 tracker representation · LaSOT | AUC: 61.9 -> 65.2 | Helped |
| Jointly encode template and search region instead of using separate streams (ablation) | MAE ViT-B/16 representation · LaSOT | AUC: 55.1 -> 44.2 | Hurt |
| Fine-tune MAE representations on tracking data without joint encoding (ablation) | MAE ViT-B/16 representation · LaSOT | AUC: 55.1 -> 50.8 | Hurt |
| Add tracking data and joint encoding while retaining the original-input reconstruction objective (ablation) | MAE ViT-B/16 representation · LaSOT | AUC: 50.8 -> 61.9 | Helped |
| Use 75% masking instead of the default 25% (ablation) | MAT · LaSOT | AUC: 65.2 -> 63.7 | Hurt |
| Use 50% masking instead of the default 25% (ablation) | MAT · LaSOT | AUC: 65.2 -> 65.6 | Mixed |
| Use 0% masking instead of the default 25% (ablation) | MAT · LaSOT | AUC: 65.2 -> 63.5 | Hurt |
| Replace random masking with grid-wise masking (ablation) | MAT · LaSOT | AUC: 63.7 -> 62.4 | Hurt |
| Replace sine-cosine positional embeddings with learnable embeddings (ablation) | MAT representation · GOT10k | AO: +LP: 69.4 | Helped |
| Do not tune the decoder (ablation) | MAT representation · GOT10k | AO: 68.0 | No clear effect |
| Remove scale jitter during pre-training (ablation) | MAT representation · LaSOT | AUC: 65.2 -> 64.5 | Hurt |
| Remove position jitter during pre-training (ablation) | MAT representation · LaSOT | AUC: 65.2 -> 57.7 | Hurt |
Closest papers by meaning
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.84)
- Compact Transformer Tracker with Correlative Masked Modeling (AAAI 2023 · similarity 0.83)
- Dual-Path Temporal Decoder for End-to-End Multi-Object Tracking (NeurIPS 2025 · similarity 0.82)
- DiffusionTrack: Diffusion Model for Multi-Object Tracking (AAAI 2024 · similarity 0.80)
- VastTrack: Vast Category Visual Object Tracking (NeurIPS 2024 · similarity 0.80)
- DeTrack: In-model Latent Denoising Learning for Visual Object Tracking (NeurIPS 2024 · similarity 0.80)
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.