Research map / Video / Video segmentation and tracking
Optical Flow in Deep Visual 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 previous frame with explicit optical flow input (Flow2s vs GOTURN) (ablation) | GOTURN · VOT2015 test | Expected Overlap: 0.1691 -> 0.1557 | No clear effect |
| add explicit optical flow to current and previous frames (Flow3s vs GOTURN) (ablation) | GOTURN · VOT2015 test | Expected Overlap: 0.1691 -> 0.1646 | No clear effect |
| use only optical flow input, dropping image frames (Flow1s) (ablation) | GOTURN · VOT2015 test | Expected Overlap: 0.1691 -> 0.0818 | Hurt |
| use lighter fully-connected branch (fewer parameters) (ablation) | GOTURN · VOT2015 test | Expected Overlap: 0.1691 -> 0.1216 | Hurt |
| use lighter fully-connected branch (ablation) | Flow2s · VOT2015 test | Expected Overlap: 0.1557 -> 0.1348 | Hurt |
| use lighter fully-connected branch (ablation) | Flow1s · VOT2015 test | Expected Overlap: 0.0818 -> 0.0829 | Helped |
| finetune flow branch (lr 1e-5) instead of fixing pretrained weights (ablation) | Flow2s · ImageNet Video validation | Overlap: 0.64 -> 0.53 | Hurt |
| finetune flow branch (lr 1e-5) instead of fixing pretrained weights (ablation) | Flow1s · ImageNet Video validation | Overlap: 0.4 -> 0.46 | Helped |
| random init of flow branch instead of action-recognition pretraining (lr 1e-6) (ablation) | Flow2s · ImageNet Video validation | Overlap: 0.64 -> 0.59 | Hurt |
Closest papers by meaning
- FlowTrack: Revisiting Optical Flow for Long-Range Dense Tracking (CVPR 2024 · similarity 0.87)
- Dense Optical Tracking: Connecting the Dots (CVPR 2024 · similarity 0.86)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.85)
- FlowFeat: Pixel-Dense Embedding of Motion Profiles (NeurIPS 2025 · similarity 0.84)
- E-MoFlow: Learning Egomotion and Optical Flow from Event Data via Implicit Regularization (NeurIPS 2025 · similarity 0.84)
- VastTrack: Vast Category Visual Object Tracking (NeurIPS 2024 · 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.