Datasets:
File size: 3,172 Bytes
12b5ac7 7b1add0 12b5ac7 | 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 62 63 64 65 66 67 68 | # Scripts
Everything used to build and verify this dataset: download, index generation,
bone-namespace normalisation, and validation.
Pure Python 3 + `requests`; no Blender or Autodesk tooling required.
## Download
Needs a Mixamo `access_token` saved in a file, passed with `--token-file`.
| script | purpose |
|---|---|
| `mixamo_download.py` | All animations as skin-free FBX on the Y Bot rig, 30 fps. Sweeps the catalogue several times because Mixamo's pagination is not stable, writes a manifest as it goes, and re-reads the token file so a long run survives a token refresh. |
| `mixamo_characters.py` | All 114 characters as T-pose FBX **with** skin mesh. |
| `mixamo_console_download.js` | Alternative route: paste into a normal Chrome console and let the browser save the files. |
```bash
python3 mixamo_download.py --pages all --skin false --token-file tok.txt --out animation
python3 mixamo_characters.py --token-file tok.txt --out character
```
## Index generation
No token needed.
| script | output |
|---|---|
| `build_prompts.py` | `../animation_prompts.json` |
| `build_frames.py` | `../animation_frames.json` |
| `build_characters.py` | `../characters.json` |
| `check_retarget.py` | `../animation_bones.json`, `../character_bones.json` |
| `build_tables.py` | `../metadata.csv`, `../characters.csv` (the browsable/loadable index) |
Two small libraries back these, and are useful on their own:
- `fbx_frames.py` — binary-FBX parser; counts keyframes. Also a CLI.
- `fbx_bones.py` — extracts real bone names from the FBX node tree. Use this rather than a
byte-level regex: `mixamorig:Hips` followed by binary data can look like a nonexistent
`mixamorig:Hipsf` bone.
## Bone-namespace fix
| script | purpose |
|---|---|
| `fbx_rename_bones.py` | Surgical binary-FBX rewriter: renames `mixamorig<N>:` → `mixamorig:` and recomputes node offsets. `--selftest <files>` asserts that a no-edit round-trip reproduces its input byte-for-byte — the correctness gate for the whole fix. |
| `refine_characters.py` | Builds `../character_refined/`: rewrites the 27 affected characters, hardlinks the other 87. |
| `verify_refined.py` | Walks the original and refined node trees in parallel and asserts every non-string property is byte-identical, so only bone names changed. |
## Verification
| script | scope |
|---|---|
| `validate.py` | Structural integrity plus index completeness and correctness. Pass a token file to add a sampled API cross-check. |
| `full_check.py` | Same, but cross-checks **every** motion name/id and character name against the Mixamo API. |
| `final_audit.py` | Widest sweep: all three FBX directories, content-duplicate detection by hash, index consistency, frame re-derivation, and animation-to-character bone compatibility. |
```bash
python3 validate.py # offline
python3 final_audit.py # full audit
```
## manifests/
Records written at download time (`file` ↔ `motion_id` / `name` / `type` / `description`).
These are the source of truth: the JSON indexes are regenerated from them plus the FBX files,
so the indexes can always be rebuilt or reformatted without re-downloading anything.
|