Research map / Detection and segmentation / Tracking
SeqTrack: Sequence to Sequence Learning for Visual Object 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 |
|---|---|---|---|
| Cast tracking as autoregressive generation of box-coordinate tokens (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 69.2 -> 64.8; 73.7 -> 72.4 | Helped |
| Encode template and search jointly rather than with separate encoders (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 62.0; 66.1 | Helped |
| Generate position before scale, using [x,y,w,h] rather than [w,h,x,y] (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 67.1; 71.8 | Helped |
| Represent boxes as center and scale rather than corner coordinates (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 68.3; 70.7 | Helped |
| Concatenate search and template features for decoder input (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 69.2 -> 69.6; 73.7 -> 73.3 | Mixed |
| Average search and template features for decoder input (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 69.2 -> 69.2; 73.7 -> 72.2 | Hurt |
| Select dynamic templates using generated-token likelihood (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 69.2 -> 69.9; 73.7 -> 76.1 | Helped |
| Update dynamic templates without reliability selection (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 69.2 -> 69.3; 73.7 -> 73.1 | Mixed |
| Use a window penalty during inference (ablation) | SeqTrack · LaSOT and GOT-10k | AUC / AO: 68.8; 73.1 | Helped |
| Add coordinates from four historical frames to the input sequence (ablation) | SeqTrack · VOT2020, GOT-10k, and TrackingNet | EAO / AO / AUC: 29.3 -> 30.1; 73.7 -> 74.9; 82.8 -> 83.0 | Helped |
| Use SeqTrack-L384 instead of the strongest reported baseline | Visual object tracking · LaSOT, LaSOText, TrackingNet, GOT-10k, TNL2K | AUC / AO: 72.5 vs 71.3; 50.7 vs 50.5; 85.5 vs 84.0; 74.8 vs 73.7; 57.8 vs 55.9 | Helped |
| Use SeqTrack-L256 instead of OSTrack on UAV123 | Visual object tracking · UAV123 | AUC: 69.7 vs 70.7 | Hurt |
Closest papers by meaning
- Autoregressive Sequential Pretraining for Visual Tracking (CVPR 2025 · similarity 0.87)
- VastTrack: Vast Category Visual Object Tracking (NeurIPS 2024 · similarity 0.86)
- DreamTrack: Dreaming the Future for Multimodal Visual Object Tracking (CVPR 2025 · similarity 0.85)
- DeTrack: In-model Latent Denoising Learning for Visual Object Tracking (NeurIPS 2024 · similarity 0.85)
- Seq2Time: Sequential Knowledge Transfer for Video LLM Temporal Grounding (CVPR 2025 · similarity 0.84)
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.84)
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.