Research map / Video / Video generation
Adaptive Capacity Autoregressive Visual Tracking
Tong Lin, Yifan Bai, Shiyi Liang, Ruigang Niu, Xing Wei
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 adaptive capacity switching between accurate and efficient trackers | ARTrack-AC · LaSOT | AUC (%) and GPU FPS: 60.2 AUC / 210 FPS (nano) and 69.4 / 65 FPS (ARTrackOT) -> 66.7 AUC / 191 FPS | Improved |
| run efficient mode in parallel instead of sequentially (ablation) | ARTrack-AC · LaSOT | GPU FPS: 131 -> 191 FPS, AUC 66.5 -> 66.7 | Improved |
| use difficulty-aware switching instead of fixed-cycle switching (ablation) | ARTrack-ACseq · LaSOT | AUC (%): 65.8 -> 66.5 | Improved |
| use supervised cosine-distance signal instead of training-free stability signal (ablation) | ARTrack-ACseq · LaSOT | AUC (%): 66.5 -> 65.9 | Hurt |
| use diffusion trajectory predictor instead of regression (ablation) | ARTrack-AC · LaSOT | AUC (%): 53.9 -> 56.9 | Improved |
| use diffusion trajectories as final output instead of localization-based tracker (ablation) | ARTrack-ACpara · LaSOT | AUC (%): 66.7 -> 56.9 (refined), 38.5 (self-conditioned) | Hurt |
| increase observation window from 5 to 10 (ablation) | ARTrack-ACpara · LaSOT | AUC (%) and GPU FPS: 66.7 AUC / 198 FPS -> 63.5 AUC / 291 FPS | Mixed |
| use predicted crop for parallel inference on all 5 frames (ablation) | ARTrack-AC · LaSOT | AUC (%) and GPU FPS: 66.5 -> 66.7 AUC; 134 -> 198 FPS | Improved |
| raise accurate-mode capacity from ARTrackOT to ARTrackv2256 (ablation) | ARTrack-ACseq · LaSOT | AUC (%): 66.5 -> 67.8 | Improved |
| replace efficient mode with non-AR HiT-Tiny (ablation) | ARTrack-ACseq · LaSOT | AUC (%): 66.5 -> 57.1 | Hurt |
| run parallel inference at 10 FPS video (GOT-10k) (ablation) | ARTrack-ACpara10 · GOT-10k | AO (%): 74.3 -> 62.6 | Hurt |
| lower video frame rate for parallel variant | ARTrack-ACpara · NFS 240 fps vs 30 fps | AUC (%): 69.4 -> 64.8 | Hurt |
Closest papers by meaning
- Autoregressive Sequential Pretraining for Visual Tracking (CVPR 2025 · similarity 0.90)
- Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion (NeurIPS 2025 · similarity 0.85)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.85)
- AR-Diffusion: Asynchronous Video Generation with Auto-Regressive Diffusion (CVPR 2025 · similarity 0.85)
- Autoregressive Video Generation without Vector Quantization (ICLR 2025 · similarity 0.85)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.85)
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.
