Add FD2 + TR and FD2 + TR + DMD checkpoints
Browse files- README.md +42 -4
- SHA256SUMS +2 -0
- checkpoints/fd2_tr_dmd_step400.ckpt +3 -0
- checkpoints/fd2_tr_step200.ckpt +3 -0
README.md
CHANGED
|
@@ -2,7 +2,7 @@
|
|
| 2 |
license: other
|
| 3 |
license_name: babel-amass-non-commercial
|
| 4 |
license_link: https://babel.is.tue.mpg.de/license.html
|
| 5 |
-
pretty_name: Triangular Resampling data
|
| 6 |
tags:
|
| 7 |
- motion-generation
|
| 8 |
- text-to-motion
|
|
@@ -12,16 +12,17 @@ tags:
|
|
| 12 |
- diffusion
|
| 13 |
---
|
| 14 |
|
| 15 |
-
# Triangular Resampling: data
|
| 16 |
|
| 17 |
-
Data for **Triangular Resampling for Long-Horizon Motion Generation**
|
| 18 |
([arXiv:2609.34697](https://arxiv.org/abs/2609.34697)).
|
| 19 |
Code: [github.com/KunhangL/TriangularResampling](https://github.com/KunhangL/TriangularResampling).
|
| 20 |
|
| 21 |
This repository holds the BABEL data and the FloodDiffusion 2 (FD2) training metadata used
|
| 22 |
in the paper. These are the exact files of the paper's runs. HumanML3D is not included; it
|
| 23 |
comes with the [FloodDiffusion release](https://huggingface.co/ShandaAI/FloodDiffusionDownloads),
|
| 24 |
-
whose `HumanML3D.zip` is identical to the HumanML3D copy used in the paper.
|
|
|
|
| 25 |
|
| 26 |
| File | Extract into | Contents | SHA-256 |
|
| 27 |
| --- | --- | --- | --- |
|
|
@@ -49,6 +50,15 @@ for name in ("BABEL.zip", "manifests.zip"):
|
|
| 49 |
zipfile.ZipFile(path).extractall("raw_data")
|
| 50 |
```
|
| 51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
## `BABEL.zip`
|
| 53 |
|
| 54 |
```
|
|
@@ -90,6 +100,31 @@ manifests/
|
|
| 90 |
`problems` field lists two HumanML3D test features with non-finite values (`007975`,
|
| 91 |
`M007975`), which FD2 validation skips.
|
| 92 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
## License
|
| 94 |
|
| 95 |
These files are derived from [BABEL](https://babel.is.tue.mpg.de/) (Punnakkal et al.,
|
|
@@ -103,6 +138,9 @@ The data is for **non-commercial scientific research only**. By using it you agr
|
|
| 103 |
[AMASS license](https://amass.is.tue.mpg.de/license.html). If you need the raw motion data or a commercial license, register
|
| 104 |
with and contact the dataset owners.
|
| 105 |
|
|
|
|
|
|
|
|
|
|
| 106 |
## Citation
|
| 107 |
|
| 108 |
```bibtex
|
|
|
|
| 2 |
license: other
|
| 3 |
license_name: babel-amass-non-commercial
|
| 4 |
license_link: https://babel.is.tue.mpg.de/license.html
|
| 5 |
+
pretty_name: Triangular Resampling data (BABEL 263D) and checkpoints
|
| 6 |
tags:
|
| 7 |
- motion-generation
|
| 8 |
- text-to-motion
|
|
|
|
| 12 |
- diffusion
|
| 13 |
---
|
| 14 |
|
| 15 |
+
# Triangular Resampling: data and checkpoints
|
| 16 |
|
| 17 |
+
Data and checkpoints for **Triangular Resampling for Long-Horizon Motion Generation**
|
| 18 |
([arXiv:2609.34697](https://arxiv.org/abs/2609.34697)).
|
| 19 |
Code: [github.com/KunhangL/TriangularResampling](https://github.com/KunhangL/TriangularResampling).
|
| 20 |
|
| 21 |
This repository holds the BABEL data and the FloodDiffusion 2 (FD2) training metadata used
|
| 22 |
in the paper. These are the exact files of the paper's runs. HumanML3D is not included; it
|
| 23 |
comes with the [FloodDiffusion release](https://huggingface.co/ShandaAI/FloodDiffusionDownloads),
|
| 24 |
+
whose `HumanML3D.zip` is identical to the HumanML3D copy used in the paper. The repository
|
| 25 |
+
also holds FD2 post-training checkpoints, described under [`checkpoints/`](#checkpoints).
|
| 26 |
|
| 27 |
| File | Extract into | Contents | SHA-256 |
|
| 28 |
| --- | --- | --- | --- |
|
|
|
|
| 50 |
zipfile.ZipFile(path).extractall("raw_data")
|
| 51 |
```
|
| 52 |
|
| 53 |
+
The checkpoints, into `checkpoints/` of the code repository:
|
| 54 |
+
|
| 55 |
+
```bash
|
| 56 |
+
hf download KunhangLI/TriangularResampling \
|
| 57 |
+
checkpoints/fd2_tr_step200.ckpt \
|
| 58 |
+
checkpoints/fd2_tr_dmd_step400.ckpt \
|
| 59 |
+
--local-dir .
|
| 60 |
+
```
|
| 61 |
+
|
| 62 |
## `BABEL.zip`
|
| 63 |
|
| 64 |
```
|
|
|
|
| 100 |
`problems` field lists two HumanML3D test features with non-finite values (`007975`,
|
| 101 |
`M007975`), which FD2 validation skips.
|
| 102 |
|
| 103 |
+
## `checkpoints/`
|
| 104 |
+
|
| 105 |
+
FD2 post-training checkpoints for HumanML3D + BABEL (263D features at 20 fps). Both start
|
| 106 |
+
from the official FD2 `humanml3d_babel_fk_200k` checkpoint
|
| 107 |
+
([AlayaLab/FloodDiffusion2](https://huggingface.co/AlayaLab/FloodDiffusion2)).
|
| 108 |
+
|
| 109 |
+
| File | Model | Config in the code repository | SHA-256 |
|
| 110 |
+
| --- | --- | --- | --- |
|
| 111 |
+
| `checkpoints/fd2_tr_step200.ckpt` | FD2 + TR, 200 post-training steps | `methods/fd2/configs/humanml_babel_tr.yaml` | `4aa7fc36cea05b6e1ac72e78bcd225a75ae6835a172e294537414545e546f104` |
|
| 112 |
+
| `checkpoints/fd2_tr_dmd_step400.ckpt` | FD2 + TR + DMD, 400 post-training steps | `methods/fd2/configs/humanml_babel_tr_dmd.yaml` | `bd505473cd8377d6c86e27c98fc573d737bfed969254bfa1708ad185a7214d16` |
|
| 113 |
+
|
| 114 |
+
- **Format.** Inference exports of about 686 MB each, with three entries: `state_dict`
|
| 115 |
+
(bfloat16), `ema_state` (float32 EMA weights, the ones generation uses) and `global_step`.
|
| 116 |
+
They hold no optimizer state, so training cannot be resumed from them.
|
| 117 |
+
- **Use.** Pass the file and its config to `methods/fd2/training/generate.py`:
|
| 118 |
+
|
| 119 |
+
```bash
|
| 120 |
+
python -m methods.fd2.training.generate --config methods/fd2/configs/humanml_babel_tr.yaml \
|
| 121 |
+
--checkpoint checkpoints/fd2_tr_step200.ckpt \
|
| 122 |
+
--manifest common/evaluation/manifests/humanml3d_prompts_256_uniform_seed1234.jsonl \
|
| 123 |
+
--output outputs/eval/fd2_tr_fixed/generated --method fd2_tr --seed 42
|
| 124 |
+
```
|
| 125 |
+
|
| 126 |
+
The code repository's README lists the 120-second evaluation scores of both checkpoints.
|
| 127 |
+
|
| 128 |
## License
|
| 129 |
|
| 130 |
These files are derived from [BABEL](https://babel.is.tue.mpg.de/) (Punnakkal et al.,
|
|
|
|
| 138 |
[AMASS license](https://amass.is.tue.mpg.de/license.html). If you need the raw motion data or a commercial license, register
|
| 139 |
with and contact the dataset owners.
|
| 140 |
|
| 141 |
+
The checkpoints are trained on HumanML3D and BABEL, starting from the official FD2 checkpoint.
|
| 142 |
+
They are released under the same terms: non-commercial scientific research only.
|
| 143 |
+
|
| 144 |
## Citation
|
| 145 |
|
| 146 |
```bibtex
|
SHA256SUMS
CHANGED
|
@@ -1,2 +1,4 @@
|
|
| 1 |
681fdd86ebc70d24bec18b1b83e04f0070a8ae8fc3cd3a33ee7f1cd7ba2dfdcf BABEL.zip
|
| 2 |
71b2b95ff2c072500562aa8ebe2b42d03dbbd4102b2d07f7e9c83e566d88cc32 manifests.zip
|
|
|
|
|
|
|
|
|
| 1 |
681fdd86ebc70d24bec18b1b83e04f0070a8ae8fc3cd3a33ee7f1cd7ba2dfdcf BABEL.zip
|
| 2 |
71b2b95ff2c072500562aa8ebe2b42d03dbbd4102b2d07f7e9c83e566d88cc32 manifests.zip
|
| 3 |
+
4aa7fc36cea05b6e1ac72e78bcd225a75ae6835a172e294537414545e546f104 checkpoints/fd2_tr_step200.ckpt
|
| 4 |
+
bd505473cd8377d6c86e27c98fc573d737bfed969254bfa1708ad185a7214d16 checkpoints/fd2_tr_dmd_step400.ckpt
|
checkpoints/fd2_tr_dmd_step400.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bd505473cd8377d6c86e27c98fc573d737bfed969254bfa1708ad185a7214d16
|
| 3 |
+
size 685790155
|
checkpoints/fd2_tr_step200.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4aa7fc36cea05b6e1ac72e78bcd225a75ae6835a172e294537414545e546f104
|
| 3 |
+
size 685775867
|