EntVista's picture
State what the free account can actually export
6fb1a0e verified
|
Raw History Blame Contribute Delete
13.3 kB
metadata
configs:
  - config_name: default
    data_files:
      - split: train
        path: metadata.jsonl
license: other
license_name: cmu-mocap-database-terms
pretty_name: CMU MoCap — De-jittered & Foot-locked (BVH)
tags:
  - 3d
  - motion-capture
  - mocap
  - animation
  - bvh
  - skeleton
  - retargeting
  - cmu
  - sample-assets
  - metrixel
size_categories:
  - n<1K

CMU MoCap — De-jittered & Foot-locked (BVH)

96 motion-capture clips from the CMU Graphics Lab Motion Capture Database, converted to BVH and processed to remove two artefacts present in the source solve: high-frequency rotational noise, and foot sliding during ground contact.

Every clip ships with a .cleanup.json alongside it recording the before/after measurements, and the headline numbers are in metadata.jsonl too — so the processing is auditable rather than asserted. Across a 21-clip validation set spanning both frame rates and 21 subjects, rotational jitter improved on 21/21 clips (median 5.3x) and in-stance foot sliding fell by 81.9%.

The motion is CMU's and remains under CMU's terms. Our contribution is the cleaning and the measurements — see Licensing below, which includes a restriction you must pass on.

Retarget these onto your own characters. Metrixel's Motion Retargeting accepts .bvh sources directly — no conversion step — and applies a whole folder of clips to a rigged character in one command. It is a Professional feature: US$59/month, or US$590/year.

A free account — 1,000 export tokens, no credit card — covers single-file runs: turntable and multi-view renders exported as PNG or JPG (wireframe included), texture extraction, and .pt mesh tensor export — so you can check output quality on your own assets before paying for the retargeting workflow.

Have your own motion library to retarget? Open a discussion on this dataset with your source format and target rig, and we can arrange Professional evaluation access.

These are motion clips only — skeletons and animation, no meshes, skins or textures. Pair them with a rigged character (see Related) to retarget motion onto a body.

Preview

What's inside

data/
├── 01_14__playground-climb-jump-down-dangle-legs-push-off-against__120fps.bvh
├── 02_07__swordplay__120fps.bvh
├── 02_08__swordplay__120fps.bvh
├── 02_09__swordplay__120fps.bvh
├── 05_02__dance-expressive-arms-pirouette__120fps.bvh
├── 05_07__dance-small-jetes-attitude-arabesque-shifted-axis-pirouette-turn__120fps.bvh
├── 05_12__dance-arms-held-high-pointe-tendue-a-terre-upper-body-rotation__120fps.bvh
├── 06_03__basketball-forward-dribble__120fps.bvh
└── … and 88 more
metadata.jsonl   one row per asset: filename + provenance + geometry stats
LICENSE.md       cmu-mocap-database-terms

Formats: .bvh ×96.

Assets

clip subject fps frames slide removed jitter removed
01_14__playground-climb-jump-down-dangle-legs-push-off-against__120fps.bvh 1 120 2338 87.3 54.8
02_07__swordplay__120fps.bvh 2 120 2251 91.2 60.2
02_08__swordplay__120fps.bvh 2 120 1500 92.6 70.4
02_09__swordplay__120fps.bvh 2 120 1033 91.1 72.8
05_02__dance-expressive-arms-pirouette__120fps.bvh 5 120 1123 91.8 70.8
05_07__dance-small-jetes-attitude-arabesque-shifted-axis-pirouette-turn__120fps.bvh 5 120 1191 66.5 69.3
05_12__dance-arms-held-high-pointe-tendue-a-terre-upper-body-rotation__120fps.bvh 5 120 1354 96.2 63.6
06_03__basketball-forward-dribble__120fps.bvh 6 120 527 85.1 72.2
06_10__basketball-forward-dribble-90-degree-left-turns__120fps.bvh 6 120 1113 94.7 64.1
06_11__basketball-forward-dribble-90-degree-right-turns__120fps.bvh 6 120 1146 93.2 66.7
06_12__basketball-forward-dribble-90-degree-right-turns-crossover-dribble__120fps.bvh 6 120 1222 93.3 54.4
09_12__navigate-walk-forward-backward-sideways__120fps.bvh 9 120 1918 91.9 65.6
104_03__frankenstein-male-walk-exact-footfalls__120fps.bvh 104 120 1619 90.3 69.7
104_10__jogstop__120fps.bvh 104 120 334 97.0 36.1
104_36__slowwalk8__120fps.bvh 104 120 2315 73.6 67.2
104_41__zombiewalk__120fps.bvh 104 120 1558 95.2 68.4
10_01__soccer-kick-ball__120fps.bvh 10 120 801 94.3 47.1
10_02__soccer-kick-ball__120fps.bvh 10 120 590 93.4 55.7
10_03__soccer-kick-ball__120fps.bvh 10 120 362 89.7 53.3
10_05__soccer-kick-ball__120fps.bvh 10 120 436 92.5 54.3
111_24__run__120fps.bvh 111 120 862 87.2 62.7
114_09__walking-up-and-down-stairs__120fps.bvh 114 120 2277 90.7 60.4
114_12__stretching__120fps.bvh 114 120 1122 98.1 68.9
135_07__mawashigeri__120fps.bvh 135 120 1472 59.0 26.8
139_20__walk-wounded-leg__120fps.bvh 139 120 1570 95.8 57.8
139_29__sneaking__120fps.bvh 139 120 2104 88.7 58.4
139_31__sneaking__120fps.bvh 139 120 2106 89.3 58.2
139_32__sneaking__120fps.bvh 139 120 1894 95.1 59.4
13_01__sit-on-high-stool-stand-up__120fps.bvh 13 120 2313 91.8 63.9
13_10__jump-up-to-grab-reach-for-tiptoe__120fps.bvh 13 120 2396 95.5 55.6
13_34__climb-ladder__120fps.bvh 13 120 1417 94.4 60.4
13_36__climb-3-steps__120fps.bvh 13 120 1379 93.2 62.4
141_13__stretch-and-yawn__120fps.bvh 141 120 368 95.7 50.7
141_14__punch-and-kick__120fps.bvh 141 120 577 86.0 38.0
143_24__kicking__120fps.bvh 143 120 1005 90.6 32.2
143_30__stretch-and-yawn__120fps.bvh 143 120 370 98.8 58.3
14_07__jump-up-to-grab-reach-for-tiptoe__120fps.bvh 14 120 2218 93.9 57.9
14_27__sit-on-high-stool-stand-up__120fps.bvh 14 120 2152 96.6 60.7
14_28__sit-on-high-stool-stand-up__120fps.bvh 14 120 2053 94.1 59.3
14_31__sit-on-stepstool-stand-up__120fps.bvh 14 120 2094 92.3 61.9
18_13__run-scramble-for-last-seat-2-subjects-subject-a__120fps.bvh 18 120 466 94.1 53.0
18_14__run-scramble-for-last-seat-2-subjects-subject-a__120fps.bvh 18 120 332 96.9 58.9
19_12__friends-meet-hang-out-a-sits-b-joins-a-2-subjects-subject-b__120fps.bvh 19 120 1253 94.5 59.8
19_13__run-scramble-for-last-seat-2-subjects-subject-b__120fps.bvh 19 120 466 97.1 57.1
21_13__blind-man-s-bluff-blindfold-tag-2-subjects-subject-b__120fps.bvh 21 120 1617 94.8 54.6
22_01__b-sits-a-pulls-up-b-2-subjects-subject-a__120fps.bvh 22 120 782 91.5 65.8
22_02__a-sits-b-pulls-up-a-2-subjects-subject-a__120fps.bvh 22 120 472 95.2 73.4
23_03__a-sits-holds-face-in-hands-b-kneels-comforts-a-2-subjects-subject-b__120fps.bvh 23 120 812 90.6 70.8
36_28__walk-on-uneven-terrain__120fps.bvh 36 120 2391 92.3 67.0
38_03__run-around-in-a-circle__120fps.bvh 38 120 761 85.6 63.3
42_01__stretch-rotate-head-shoulders-arms-legs__60fps.bvh 42 60 1134 95.6 62.3
43_02__playground-grip-bar-swing-body__120fps.bvh 43 120 904 80.4 72.0
43_03__playground-grip-bar-swing-body__120fps.bvh 43 120 840 98.2 70.4
49_02__jump-up-and-down-hop-on-one-foot__120fps.bvh 49 120 2085 90.1 50.1
49_04__run-leap__120fps.bvh 49 120 251 90.8 48.3
49_16__dance-fold-in-from-side-arabesque-curl-inwards__120fps.bvh 49 120 1136 98.7 81.1
49_20__balance-on-one-leg-outstretched-arms__120fps.bvh 49 120 898 84.6 69.0
54_14__monkey-human-subject__120fps.bvh 54 120 1628 88.2 49.8
55_01__dance-whirl__120fps.bvh 55 120 1806 96.1 62.7
55_02__lambada-dance__120fps.bvh 55 120 2180 95.8 64.4
60_10__salsa-dance__60fps.bvh 60 60 1197 94.1 30.6
61_10__salsa-dance__60fps.bvh 61 60 1197 92.4 45.0
62_19__opening-a-box__60fps.bvh 62 60 659 96.9 30.5
63_01__swing__120fps.bvh 63 120 448 99.2 56.5
64_01__swing__120fps.bvh 64 120 448 99.2 56.5
64_06__swing__120fps.bvh 64 120 529 99.8 45.8
64_07__swing__120fps.bvh 64 120 376 99.9 44.4
64_09__swing__120fps.bvh 64 120 452 99.7 45.9
74_06__kick__60fps.bvh 74 60 325 95.3 40.0
75_07__angle-jumps__60fps.bvh 75 60 1119 97.1 41.9
75_08__360-jumps__60fps.bvh 75 60 1115 97.7 36.2
76_08__careful-stepping-over-things__120fps.bvh 76 120 1503 91.7 75.7
77_10__careful-run__60fps.bvh 77 60 690 87.6 44.3
77_12__careful-run-in-circle__60fps.bvh 77 60 978 89.7 53.9
77_14__careful-creeping__60fps.bvh 77 60 990 95.6 63.4
77_15__careful-creeping__60fps.bvh 77 60 1078 95.3 71.4
79_02__swimming__60fps.bvh 79 60 776 87.6 59.9
79_17__playing-violin__60fps.bvh 79 60 804 90.7 58.8
79_44__washing-a-window__60fps.bvh 79 60 854 92.0 62.6
79_55__sweeping__60fps.bvh 79 60 640 97.9 60.8
81_06__push-heavy-box__120fps.bvh 81 120 1274 87.0 34.8
81_14__climb-backwards-off-ledge__120fps.bvh 81 120 2090 98.5 70.5
81_15__climb-backwards-off-ledge__120fps.bvh 81 120 2027 96.1 55.0
82_03__jump-off-ledge__120fps.bvh 82 120 1899 92.9 61.2
82_04__jump-off-ledge__120fps.bvh 82 120 908 95.0 49.1
82_08__stand-still-casual-walk-forward__120fps.bvh 82 120 1699 94.2 72.9
85_06__kickflip__120fps.bvh 85 120 950 95.3 25.1
85_07__kickflipstumble__120fps.bvh 85 120 959 95.1 27.7
87_05__cartwheels__60fps.bvh 87 60 511 65.7 41.1
88_05__cartwheel-into-backflip__60fps.bvh 88 60 149 88.5 35.3
88_07__cartwheel__60fps.bvh 88 60 156 91.3 34.6
88_09__stretch-and-cartwheel__60fps.bvh 88 60 376 77.5 45.1
88_10__stretch-and-spin__60fps.bvh 88 60 467 90.1 35.9
90_02__cartwheel__120fps.bvh 90 120 781 88.1 44.2
90_03__cartwheel__120fps.bvh 90 120 884 87.8 45.2
90_04__cartwheel__120fps.bvh 90 120 698 84.1 38.4

Licensing

The motion originates from the CMU Graphics Lab Motion Capture Database and is used under CMU's published terms. EntVista Studio does not own it and grants no licence over it.

  • Free for all uses, including research and commercially-sold products.
  • You may not resell the motion data itself, and CMU states this explicitly covers converted form — BVH, FBX, GLB or retargeted onto another character. Shipping a product containing it is fine; selling the data is not.
  • This restriction transfers to you and to anyone you redistribute to.

If you publish work using this data, CMU asks that you include: The data used in this project was obtained from mocap.cs.cmu.edu. The database was created with funding from NSF EIA-0196217.

Source archive was the official allasfamc.zip bulk download; the database was not crawled.

Full terms in LICENSE.md. Read it before use — CMU permits free use and inclusion in commercially-sold products, but prohibits reselling the data itself even in converted form, and that restriction passes to you.

Honest limits

  • The motion is not ours and the cleaning is not magic. Processing reduces jitter and foot sliding; it does not reconstruct missing markers, fix a bad solve, or add detail that was never captured.
  • Foot locking assumes ground contact is detectable from contact-point speed. It is skipped where that fails — acrobatics, ladders, and motion with no clear stance phase.
  • A small pelvis give is applied (capped at 4% of body height, low-passed at 2 Hz) where a foot pin would otherwise be out of reach. The root's horizontal trajectory is never modified.
  • CMU skeletons vary by subject, including genuine left/right asymmetry. Clips are not normalised to a common rig — retarget rather than assuming a shared skeleton.
  • These 96 are a curated subset, selected for measured improvement and motion variety, not a representative sample of CMU's 2500+ trials.

Related