List your company

Point Tracking on RoboTAP

49AJTAPTR
48.03254.56661.167.634Feb 4, 2026Feb 10, 2026Feb 16, 2026Feb 22, 2026Feb 28, 2026Mar 6, 2026Mar 12, 2026

Evaluation results

MethodLinks
TAPTRTrain data=Kub, Track Type=Sparse2026.024964.485.2—
TAPIRTrain data=Kub, Track Type=Sparse2026.0249.664.285—
CoTrackerTrain data=Kub, Track Type=Sparse2026.0249.664.383.3—
LocoTrackTrain data=Kub, Track Type=Sparse2026.0252.966.885.3—
BootsTAPIRTrain data=Kub+15M, Track Type=Sparse2026.0254.668.486.5—
CoTracker3Train data=Kub+15k, Track Type=Sparse, Inference Mode=offline2026.0254.767.887.4—
CoTracker3Train data=Kub+15k, Track Type=Sparse, Inference Mode=online2026.0255.868.588.3—
TAPNext-BTraining Strategy=Synthetic Pretraining2026.0359.5—8872.8
TAPIRTraining Strategy=Synthetic Pretraining2026.0359.6—8773.4
DELTATrain data=Kub, Track Type=Dense2026.0260.273.485.6—
CoTracker3 (Window)Training Strategy=Synthetic Pretraining2026.0360.8—87.173.7
LocoTrackTraining Strategy=Synthetic Pretraining2026.0362.3—87.176.2
BootsTAPNext-BTraining Strategy=Real-World Fine-Tuning2026.0364—88.775
Anthro-LocoTrackTraining Strategy=Real-World Fine-Tuning2026.0364.7—88.479.2
BootsTAPIRTraining Strategy=Real-World Fine-Tuning2026.0364.9—86.380.1
CoTracker3 (Window)Training Strategy=Real-World Fine-Tuning2026.0366.4—90.878.8
Track-On2Training Strategy=Synthetic Pretraining2026.0368.1—93.480.5
AllTrackerTraining Strategy=Real-World Fine-Tuning, Leverages additional optical-flow=true2026.0368.8—92.280.9
AllTrackerTrain data=Kub, Track Type=Dense2026.0270.181.592.2—
Track-On-RTraining Strategy=Real-World Fine-Tuning2026.0370.9—9482.6
AllTrackerTrain data=Kub+Mix, Track Type=Dense2026.0271.983.492.8—
CoWTrackerTrain data=Kub, Track Type=Dense2026.0273.283.494.7—
AccFlowzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03—57.9——
MegaFlowzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03—77.5——
MemFlow-Tzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03—77——
PIPs++Train data=PO, Track Type=Sparse2026.02—63.5——
RAFTzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03—72.2——
SEA-RAFTzero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03—67.6——
WAFT-DINOv3-a2zero-shot=true, input resolution=384 × 512, trained on=optical flow datasets exclusively2026.03—73——