Models

Tracking models

12 tracking models. Each section says what the model is, whether it may be used commercially and how fast it is; licences and checks are under Details.

Input: a video (its frames in order) and query points (t, x, y). Output: the position of every query in every frame and the probability that it is visible there.

Measured on RTX 5060 Ti, torch 2.14.1+cu130; fastforks bench MODEL --size HxW measures yours. The summary uses one input, 480x854: kernels time with its speed-up over the faster original run, and kernels peak reserved memory with its ratio to the smaller original one. Where measured (one variant per family), a variant's table also lists portable under torch.compile and the fp16 TensorRT engine of portable (export; memory: the engine's device memory + torch's peak). – = not measured; no kernels = the family has no kernels backend (reason under its table). Reserved memory is the allocator's cache: kernels (CUDA graphs, cuDNN workspaces) can reserve a few percent to some tens of percent more than portable at the same allocated peak.

Other devices: Apple M1 Pro 16 GB, NVIDIA L4, NVIDIA L40S, NVIDIA A10, NVIDIA A100-SXM4-40GB, NVIDIA RTX PRO 6000; across devices.

modelcommercialkernels time (ms),
speed-up vs original
kernels peak memory (MiB, reserved),
ratio to original
alltrackeryes221, 2.2x1288, 0.56x
alltracker-tinyyes217, 2.0x1166, 0.53x
bootstapiryes192, 1.3x628, 0.81x
cotracker3-offlineno154, 1.6x882, 0.36x
cotracker3-onlineno176, 1.9x888, 0.47x
locotrack-byes173, 1.2x660, 0.84x
locotrack-syes151, 1.3x642, 0.79x
tapiryes141, 1.3x548, 0.95x
tapnextyes381, 2.4x1254, 0.66x
tapnext-ppyes381, 3.9x1254, 0.73x
track-on-ryes1102, 1.8x1620, 0.84x
track-on2yes1108, 1.8x1620, 1.0x

All models

Modes, licences, training data and outputs per model

Commercial status per variant: yes if the code and the weights licence both allow commercial use (the training data is listed but does not change it; weights without a licence of their own are assumed to carry the code licence); no if either forbids it; unclear for custom licence terms (treat as no).

modelmodescode licenceweights licencetraining datacommercialoutputs
alltrackerportable, kernelsMITMITsceneflow, flyingchairs, spring, hd1k, kitti, tartanair, unstatedyestracks, visibility
alltracker-tinyportable, kernelsMITMITsceneflow, flyingchairs, spring, hd1k, kitti, tartanair, unstatedyestracks, visibility
bootstapirportable, kernelsApache-2.0Apache-2.0unstatedyestracks, visibility
cotracker3-offlineportable, kernelsCC-BY-NC-4.0CC-BY-NC-4.0unstatedno (code: CC-BY-NC-4.0 (noncommercial); weights: CC-BY-NC-4.0)tracks, visibility
cotracker3-onlineportable, kernelsCC-BY-NC-4.0CC-BY-NC-4.0unstatedno (code: CC-BY-NC-4.0 (noncommercial); weights: CC-BY-NC-4.0)tracks, visibility
locotrack-bportable, kernelsApache-2.0Apache-2.0unstatedyestracks, visibility
locotrack-sportable, kernelsApache-2.0Apache-2.0unstatedyestracks, visibility
tapirportable, kernelsApache-2.0Apache-2.0unstatedyestracks, visibility
tapnextportable, kernelsApache-2.0Apache-2.0unstatedyestracks, visibility
tapnext-ppportable, kernelsApache-2.0Apache-2.0unstatedyestracks, visibility
track-on-rportable, kernelsMIT AND LicenseRef-DINOv3-LicenseMIT AND LicenseRef-DINOv3-Licenseunstatedyestracks, visibility
track-on2portable, kernelsMIT AND LicenseRef-DINOv3-LicenseMIT AND LicenseRef-DINOv3-Licenseunstatedyestracks, visibility

Training datasets

iddatasetlicencecommercial use
sceneflowFreiburg SceneFlow (FlyingThings3D, Driving, Monkaa)LicenseRef-research-onlyno
kittiKITTI stereo / optical flow 2012 / 2015CC-BY-NC-SA-3.0no
tartanairTartanAirCC-BY-4.0yes
flyingchairsFlyingChairs (Freiburg)LicenseRef-research-onlyno
hd1kHCI HD1K optical flow benchmarknot statedunclear
springSpring (scene flow / optical flow / stereo, University of Stuttgart)CC-BY-4.0yes