Point Tracking on TAP-Vid Kinetics
90.5Overall AccuracyTrack-On-R
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Track-On-RTraining Strategy=Real-World Fine-Tuning2026.03 | 90.5 | 57.8 | 71 | |
| TAPNextSupervision=Supervised2025.12 | 90.06 | 54.94 | — | |
| Track-On2Training Strategy=Synthetic Pretraining2026.03 | 89.6 | 55.3 | 69.3 | |
| CoTracker3Supervision=Supervised2025.12 | 89.43 | 54.35 | — | |
| DINTR2024.10 | 89.4 | 57.8 | 72.5 | |
| AllTrackerTraining Strategy=Real-World Fine-Tuning, Leverages additional optical-flow=true2026.03 | 89.1 | 56.8 | 69.3 | |
| CoTracker3 (Window)Training Strategy=Real-World Fine-Tuning2026.03 | 88.3 | 55.8 | 68.5 | |
| TAPIR2024.10 | 87.6 | 57.1 | 70 | |
| CoTracker3†Training Dataset=Kub. + 15k, Steps=65k, Sequence Length=60-80, Batch Size=322025.12 | 87.5 | 54.7 | 67.8 | |
| BootsTAPNext-BTraining Strategy=Real-World Fine-Tuning2026.03 | 87.4 | 57.3 | 70.6 | |
| CoTracker3 (Window)Training Strategy=Synthetic Pretraining2026.03 | 87.1 | 54.1 | 66.6 | |
| TAPNext-BTraining Strategy=Synthetic Pretraining2026.03 | 87 | 53.3 | 67.9 | |
| CoTracker-SingleEvaluation Mode=First, Input Resolution=256x2562024.07 | 86.5 | 48.7 | 64.3 | |
| BootsTAPIRTraining Dataset=Kub. + 15M, Steps=200k, Sequence Length=24, Batch Size=1,5362025.12 | 86.5 | 54.6 | 68.4 | |
| BootsTAPIRTraining Strategy=Real-World Fine-Tuning2026.03 | 86.5 | 54.6 | 68.4 | |
| Anthro-LocoTrackTraining Strategy=Real-World Fine-Tuning2026.03 | 86.4 | 53.9 | 68.4 | |
| BootsTAP+Evaluation Mode=First, Input Resolution=256x256, Extra Training Data=15M video clips2024.07 | 86.3 | 54.7 | 68.5 | |
| TAPTRv2Evaluation Mode=First, Input Resolution=256x256, Backbone=ResNet-502024.07 | 85.7 | 49.7 | 64.2 | |
| LocoTrackTraining Strategy=Synthetic Pretraining2026.03 | 85.3 | 52.9 | 66.8 | |
| TAPTREvaluation Mode=First, Input Resolution=256x256, Backbone=ResNet-502024.07 | 85.2 | 49 | 64.4 | |
| TAPTRv2Training Dataset=Kub., Steps=44k, Sequence Length=-, Batch Size=322025.12 | 85.2 | 49 | 64.4 | |
| TAP-Net2024.10 | 85 | 46.6 | 60.9 | |
| TAPIREvaluation Mode=First, Input Resolution=256x2562024.07 | 85 | 49.6 | 64.2 | |
| TAPIRTraining Dataset=Kub., Steps=50k, Sequence Length=24, Batch Size=42025.12 | 85 | 49.6 | 64.2 | |
| TAPIRTraining Strategy=Synthetic Pretraining2026.03 | 85 | 49.6 | 64.2 | |
| DiTrackerTraining Dataset=Kub., Steps=36k, Sequence Length=46, Batch Size=42025.12 | 84.5 | 54.3 | 67.4 | |
| Opt-CWMSupervision=Self-Supervised2025.12 | 84.12 | 44.85 | — | |
| TAPIRSupervision=Supervised2025.12 | 82.16 | 45.58 | — | |
| LocoTrackTraining Dataset=Kub., Steps=400k, Sequence Length=24-48, Batch Size=82025.12 | 82.1 | 52.3 | 66.4 | |
| TAP-NetSupervision=Supervised2025.12 | 80.88 | 34.59 | — | |
| TAP-NetEvaluation Mode=First, Input Resolution=256x2562024.07 | 80.6 | 38.5 | 54.4 | |
| Kubric-VFS-Like2024.10 | 80 | 40.5 | 59 | |
| Context-PIPsEvaluation Mode=First, Input Resolution=256x2562024.07 | 79.8 | — | — | |
| RAFT2024.10 | 79.7 | 34.5 | 52.5 | |
| CoTracker3Training Dataset=Kub., Steps=50k, Sequence Length=60, Batch Size=322025.12 | 78.7 | 49.3 | 62.7 | |
| HeFTSupervision=Zero-Shot, Backbone=Cosmos-Predict2-2B2025.12 | 78.6 | 36.64 | — | |
| PIPs2024.10 | 77.1 | 35.1 | 54.8 | |
| HeFTSupervision=Zero-Shot, Backbone=Wan2.1-1.3B2025.12 | 73.6 | 30 | — | |
| PIPsEvaluation Mode=First, Input Resolution=256x2562024.07 | 72.9 | 31.7 | 53.7 | |
| MFTEvaluation Mode=First, Input Resolution=256x2562024.07 | 72.7 | 39.6 | 60.4 | |
| Point-PromptingSupervision=Zero-Shot2025.12 | 71.39 | 27.36 | — | |
| GMRWSupervision=Self-Supervised2025.12 | 71.33 | 25.7 | — | |
| HeFTSupervision=Zero-Shot, Backbone=CogvideoX-2B2025.12 | 66.22 | 17.29 | — | |
| DIFTSupervision=Zero-Shot2025.12 | 63.17 | 15.1 | — | |
| SD-DINOSupervision=Zero-Shot2025.12 | 62.79 | 16.47 | — | |
| DINOv2+NNSupervision=Zero-Shot2025.12 | 62.45 | 12.69 | — | |
| RAFTSupervision=Supervised2025.12 | 59.78 | 21.52 | — | |
| COTR2024.10 | 57.4 | 19 | 38.8 | |
| AccFlowzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03 | — | — | 38.8 | |
| MegaFlowzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03 | — | — | 65.5 | |
| MemFlow-Tzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03 | — | — | 64.9 | |
| RAFTzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03 | — | — | 64.3 | |
| SEA-RAFTzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03 | — | — | 64.3 | |
| WAFT-DINOv3-a2zero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03 | — | — | 61 |