File size: 2,630 Bytes
58ee0a1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
{
  "schema": 1,
  "release": "kimore-v2-20260909",
  "status": "official",
  "approval": "User approved reconstruction, synthetic motion appearance, data and promotion to the official approach after reviewing the KIMORE benchmark.",
  "config": "MotionGPT3/configs/rehab_official.yaml",
  "generator": {
    "path": "checkpoints/official/kimore_v2/generator.ckpt",
    "source_within_experiment": "generator_v2/best_validation.ckpt",
    "sha256": "46ad3b39607eaa4083ce107b9c5734fe2e8984f8256182bd8501b14467b814cf",
    "epoch": 40,
    "updates": 4920,
    "completed_training_epochs": 100,
    "completed_training_updates": 12300,
    "selection": "lowest exercise-macro validation diffusion loss over the evaluated trained checkpoints"
  },
  "vae": {
    "path": "checkpoints/official/kimore_v2/vae.ckpt",
    "source_within_experiment": "vae_package/vae.ckpt",
    "sha256": "446de7711e4cdff79ec8e08ceb15161f10e4293785462397d04397e633ddf46e",
    "reconstruction_checkpoint_sha256": "2daf07ad6d24c7decedc42fe4f24be5e23e94a9b8f4138347a5fea8a60b88367",
    "selection": "visually reviewed encoder-only refinement at 10000 additional updates",
    "latent_shape": [1, 256],
    "latent_target": "mean",
    "latent_inverse_std": 0.6275949456609311,
    "frozen_during_generator_training": true,
    "posterior_variance_calibrated": false
  },
  "data": {
    "root": "data/rehab_timing_v2",
    "conversion_manifest_sha256": "dd79a30a10141056f3e36dbc88ddd3c90f459280b84e3f1603225887ef64da8e",
    "mean_sha256": "ba24d49c8044195751b3b49ba57d25e2e2ae75b27804a57c15eb627bb5590495",
    "std_sha256": "7982df965c8a4751d53a85488de77468f83a50794f0e6bdf5f69bbd57f3320f8",
    "fps": 20,
    "joints": 22,
    "features": 263,
    "eligible_generator_train_clips": 1965,
    "generator_train_subjects": 55
  },
  "generation": {
    "skeleton": "none",
    "rigidification": false,
    "coordinate_frame": "HumanML3D canonical frame",
    "guidance_scale": 3.0,
    "temperature": 1.0,
    "duration_source": "training split only",
    "kinematic_filtering": false
  },
  "benchmark": {
    "results": "results/generator_refinement_20260909/kimore_benchmark",
    "n_real": [2, 5, 10, 20, 55],
    "architectures": ["CNN 1D", "ST-GCN"],
    "synthetic_clips_per_arm": 1000,
    "classifier_seeds": [0, 1, 2],
    "input_orientation": "180-degree yaw on separate synthetic classifier copies",
    "scope": "n_real limits classifier subjects; generator retains access to all 55 training subjects"
  },
  "experiment": "results/generator_refinement_20260909",
  "documentation": "docs/experiments/2026-09-09_official-vae-generator.md"
}