docs: refer to the validation clip generically as a reference walk
Browse files
README.md
CHANGED
|
@@ -27,7 +27,7 @@ Generative text→skeletal-animation model shipped with
|
|
| 27 |
|
| 28 |
Earlier versions trained on the **CMU MoCap** corpus and inherited a facing
|
| 29 |
problem: generated walks moved convincingly but oriented backwards. Measuring
|
| 30 |
-
against a real
|
| 31 |
away from it, *further than a real punch, dance or jump*, and the real-vs-real
|
| 32 |
floor is 0.437. The corpus never carried the correct convention, so the model
|
| 33 |
could not learn it. v6.x models scored 1.89 while their own training data
|
|
@@ -46,7 +46,7 @@ new loss:
|
|
| 46 |
ankle. The v7.x walk bent the knee 25° and the ankle 31° (real: 38°/10°),
|
| 47 |
which renders as a near-straight knee with a false second joint below it.
|
| 48 |
|
| 49 |
-
### Measured against a real
|
| 50 |
|
| 51 |
| | v7.x | **v8.0** | real |
|
| 52 |
|---|---|---|---|
|
|
|
|
| 27 |
|
| 28 |
Earlier versions trained on the **CMU MoCap** corpus and inherited a facing
|
| 29 |
problem: generated walks moved convincingly but oriented backwards. Measuring
|
| 30 |
+
against a real reference walk showed why — CMU "walk" windows sit 1.73–2.07 rad
|
| 31 |
away from it, *further than a real punch, dance or jump*, and the real-vs-real
|
| 32 |
floor is 0.437. The corpus never carried the correct convention, so the model
|
| 33 |
could not learn it. v6.x models scored 1.89 while their own training data
|
|
|
|
| 46 |
ankle. The v7.x walk bent the knee 25° and the ankle 31° (real: 38°/10°),
|
| 47 |
which renders as a near-straight knee with a false second joint below it.
|
| 48 |
|
| 49 |
+
### Measured against a real reference walk
|
| 50 |
|
| 51 |
| | v7.x | **v8.0** | real |
|
| 52 |
|---|---|---|---|
|