Pose models
47 pose 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: one RGB image and a box per person. Output: keypoints with confidence scores for every person.
Measured on RTX 5060 Ti, torch 2.14.1+cu130; fastforks bench MODEL --size HxW measures yours. The summary uses one input, 1080x1920: 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.
| model | commercial | kernels time (ms), speed-up vs original | kernels peak memory (MiB, reserved), ratio to original |
|---|---|---|---|
dwpose-l | yes | 6.20, 28.7x | 266, 0.69x |
dwpose-l-384 | yes | 12.7, 15.2x | 404, 0.5x |
dwpose-m | yes | 4.86, 33.8x | 200, 0.25x |
dwpose-s | yes | 3.35, 51.6x | 178, 0.21x |
dwpose-t | yes | 3.17, 50.6x | 174, 0.2x |
hrnet-w32 | yes | 9.75, 18.0x | 194, 0.6x |
hrnet-w32-384 | yes | 19.7, 9.7x | 294, 0.39x |
hrnet-w48 | yes | 15.6, 12.8x | 262, 0.57x |
hrnet-w48-384 | yes | 34.2, 7.1x | 360, 0.39x |
retinaface-mobilenet0.25 | yes | 2.74, 27.7x | 128, 1.2x |
retinaface-resnet50 | yes | 11.7, 6.8x | 464, 0.95x |
rf-detr-keypoint-preview | yes | 9.46, 2.0x | 224, 0.92x |
rtmo-l | yes | 8.30, 2.1x | 228, 0.35x |
rtmo-m | yes | 6.85, 2.1x | 212, 0.59x |
rtmo-s | yes | 6.27, 1.8x | 180, 0.82x |
rtmo-t | yes | 6.31, 1.8x | 156, 1.4x |
rtmpose-l | yes | 5.70, 28.3x | 244, 0.22x |
rtmpose-l-384 | yes | 11.4, 14.9x | 390, 0.95x |
rtmpose-l-halpe26 | yes | 5.69, 28.9x | 244, 0.46x |
rtmpose-l-halpe26-384 | yes | 11.5, 18.0x | 410, 0.39x |
rtmpose-m | yes | 4.50, 39.7x | 200, 0.55x |
rtmpose-m-384 | yes | 9.23, 18.1x | 320, 0.55x |
rtmpose-m-ap10k | yes | 5.42, 34.4x | 200, 0.25x |
rtmpose-m-face6 | yes | 5.77, 34.4x | 222, 0.32x |
rtmpose-m-halpe26 | yes | 4.53, 39.7x | 180, 0.57x |
rtmpose-m-halpe26-384 | yes | 9.28, 20.7x | 320, 0.42x |
rtmpose-m-hand5 | yes | 5.46, 35.2x | 200, 0.42x |
rtmpose-s | yes | 3.08, 52.1x | 158, 0.33x |
rtmpose-s-face6 | yes | 3.74, 63.4x | 202, 0.37x |
rtmpose-s-halpe26 | yes | 3.05, 56.3x | 178, 0.89x |
rtmpose-t | yes | 2.90, 52.0x | 156, 0.32x |
rtmpose-t-face6 | yes | 3.61, 52.1x | 180, 0.39x |
rtmpose-t-halpe26 | yes | 2.90, 59.2x | 156, 1.1x |
rtmpose-x-halpe26-384 | yes | 19.1, 11.4x | 448, 0.81x |
rtmw-l | yes | 9.99, 20.3x | 312, 0.44x |
rtmw-l-384 | yes | 20.6, 13.3x | 458, 0.45x |
rtmw-m | yes | 7.54, 22.5x | 266, 0.81x |
rtmw-x | yes | 14.4, 14.3x | 392, 0.62x |
rtmw-x-384 | yes | 50.9, 5.7x | 632, 0.35x |
vitpose-base | yes | 12.0, 8.8x | 364, 0.6x |
vitpose-base-simple | yes | 12.1, 9.0x | 344, 0.49x |
vitpose-plus-base | yes | 11.9, 9.5x | 344, 0.46x |
vitpose-plus-huge | yes | 62.5, 3.9x | 1406, 0.34x |
vitpose-plus-large | yes | 33.2, 5.0x | 764, 0.36x |
vitpose-plus-small | yes | 6.05, 17.2x | 198, 0.64x |
yunet-n | yes | 2.22, 2.7x | 104, 0.62x |
yunet-s | yes | 2.00, 2.8x | 104, 0.7x |
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).
| model | modes | code licence | weights licence | training data | commercial | outputs |
|---|---|---|---|---|---|---|
dwpose-l | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, coco, aic, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
dwpose-l-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, coco, aic, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
dwpose-m | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, coco, aic, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
dwpose-s | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, coco, aic, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
dwpose-t | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, coco, aic, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
hrnet-w32 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
hrnet-w32-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
hrnet-w48 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
hrnet-w48-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
retinaface-mobilenet0.25 | portable, kernels | MIT | MIT | wider-face | yes | keypoints, keypoint_scores, boxes, scores |
retinaface-resnet50 | portable, kernels | MIT | MIT | wider-face | yes | keypoints, keypoint_scores, boxes, scores |
rf-detr-keypoint-preview | portable, kernels | Apache-2.0 | Apache-2.0 | coco, lvd-142m | yes | keypoints, keypoint_scores, boxes, scores |
rtmo-l | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18 | yes | keypoints, keypoint_scores, boxes, scores |
rtmo-m | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18 | yes | keypoints, keypoint_scores, boxes, scores |
rtmo-s | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18 | yes | keypoints, keypoint_scores, boxes, scores |
rtmo-t | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18 | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-l | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-l-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-l-halpe26 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-l-halpe26-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-m | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-m-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-m-ap10k | portable, kernels | Apache-2.0 | Apache-2.0 | ap-10k, coco, aic, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-m-face6 | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, halpe, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-m-halpe26 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-m-halpe26-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-m-hand5 | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, halpe, coco, aic, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-s | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-s-face6 | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, halpe, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-s-halpe26 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-t | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-t-face6 | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, halpe, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-t-halpe26 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmpose-x-halpe26-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmw-l | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmw-l-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmw-m | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmw-x | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
rtmw-x-384 | portable, kernels | Apache-2.0 | Apache-2.0 | coco-wholebody, ubody, aic, crowdpose, mpii, jhmdb, halpe, posetrack18, coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
vitpose-base | portable, kernels | Apache-2.0 | Apache-2.0 | coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
vitpose-base-simple | portable, kernels | Apache-2.0 | Apache-2.0 | coco, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
vitpose-plus-base | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, mpii, ap-10k, apt-36k, coco-wholebody, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
vitpose-plus-huge | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, mpii, ap-10k, apt-36k, coco-wholebody, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
vitpose-plus-large | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, mpii, ap-10k, apt-36k, coco-wholebody, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
vitpose-plus-small | portable, kernels | Apache-2.0 | Apache-2.0 | coco, aic, mpii, ap-10k, apt-36k, coco-wholebody, imagenet-1k | yes | keypoints, keypoint_scores, boxes, scores |
yunet-n | portable, kernels | BSD-3-Clause | BSD-3-Clause | wider-face | yes | keypoints, keypoint_scores, boxes, scores |
yunet-s | portable, kernels | BSD-3-Clause | BSD-3-Clause | wider-face | yes | keypoints, keypoint_scores, boxes, scores |
Training datasets
| id | dataset | licence | commercial use |
|---|---|---|---|
coco | COCO 2017 (train, instance annotations) | CC-BY-4.0 | unclear |
imagenet-1k | ImageNet-1k (ILSVRC 2012 classification) | LicenseRef-noncommercial | no |
lvd-142m | LVD-142M (DINOv2's curated web-image set, not released) | not stated | unclear |
aic | AI Challenger human keypoints (2017) | not stated | unclear |
mpii | MPII Human Pose | BSD-2-Clause | no |
ap-10k | AP-10K (animal pose) | CC-BY-4.0 | unclear |
apt-36k | APT-36K (animal pose tracking) | MIT | unclear |
coco-wholebody | COCO-WholeBody | CC-BY-NC-4.0 | no |
ubody | UBody (upper-body whole-body keypoints, from OSX) | not stated | unclear |
crowdpose | CrowdPose | not stated | unclear |
halpe | Halpe-FullBody | not stated | unclear |
posetrack18 | PoseTrack 2018 | not stated | unclear |
jhmdb | sub-JHMDB (J-HMDB) | not stated | unclear |
wider-face | WIDER FACE (32k images, 394k labelled faces) | LicenseRef-research-only | no |