m-hamza-mughal's picture
Upload folder using huggingface_hub
c0fa758 verified
Raw
History Blame Contribute Delete
13 kB
data_keys = [
'motion', 'motion_mask', 'motion_length', 'melspec', 'audio', 'text',
'gest_types', 'discourse', 'speaker_id'
]
train_pipeline = []
motion_length = 150
motion_fps = 15
audio_sr = 16000
base_data_cfg = dict(
type='BEATXDataset',
pose_rep='smplxflame_30',
facial_rep='smplxflame_30',
data_path=
'/CT/GestureSynth1/work/GestureGPT/PantoMatrix/BEAT2/beat_english_v2.0.0/',
cache_path=
'./cache/beatx_cache_backup/',
debug=False,
tiny=False,
face_joint_idx=[18, 13, 9, 5],
sample_rate=16000,
num_mels=80,
hop_length=512,
fps=15,
stride=5,
pose_length=150,
ori_joints='beat_smplx_joints',
deps_path='/CT/GestureSynth1/work/GestureGPT/GestureRep/deps/',
training_speakers=[
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
],
additional_data=True,
pose_fps=15,
audio_sr=16000,
audio_fps=16000,
new_cache=False,
clean_first_seconds=0,
clean_final_seconds=0,
audio_rep='wav2vec',
word_rep='bert_framealigned',
id_rep='idx',
sem_rep='info',
prom_rep='prom',
emo_rep='emo')
train_cfg = dict(
type='BEATXDataset',
pose_rep='smplxflame_30',
facial_rep='smplxflame_30',
data_path=
'/CT/GestureSynth1/work/GestureGPT/PantoMatrix/BEAT2/beat_english_v2.0.0/',
cache_path=
'./cache/beatx_cache_backup/',
debug=False,
tiny=False,
face_joint_idx=[18, 13, 9, 5],
sample_rate=16000,
num_mels=80,
hop_length=512,
fps=15,
stride=5,
pose_length=150,
ori_joints='beat_smplx_joints',
deps_path='/CT/GestureSynth1/work/GestureGPT/GestureRep/deps/',
training_speakers=[
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
],
additional_data=True,
pose_fps=15,
audio_sr=16000,
audio_fps=16000,
new_cache=False,
clean_first_seconds=0,
clean_final_seconds=0,
audio_rep='wav2vec',
word_rep='bert_framealigned',
id_rep='idx',
sem_rep='info',
prom_rep='prom',
emo_rep='emo',
split='train')
val_cfg = dict(
type='BEATXDataset',
pose_rep='smplxflame_30',
facial_rep='smplxflame_30',
data_path=
'/CT/GestureSynth1/work/GestureGPT/PantoMatrix/BEAT2/beat_english_v2.0.0/',
cache_path=
'./cache/beatx_cache_backup/',
debug=False,
tiny=False,
face_joint_idx=[18, 13, 9, 5],
sample_rate=16000,
num_mels=80,
hop_length=512,
fps=15,
stride=5,
pose_length=150,
ori_joints='beat_smplx_joints',
deps_path='/CT/GestureSynth1/work/GestureGPT/GestureRep/deps/',
training_speakers=[
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
],
additional_data=True,
pose_fps=15,
audio_sr=16000,
audio_fps=16000,
new_cache=False,
clean_first_seconds=0,
clean_final_seconds=0,
audio_rep='wav2vec',
word_rep='bert_framealigned',
id_rep='idx',
sem_rep='info',
prom_rep='prom',
emo_rep='emo',
split='val')
test_cfg = dict(
type='BEATXDataset',
pose_rep='smplxflame_30',
facial_rep='smplxflame_30',
data_path=
'/CT/GestureSynth1/work/GestureGPT/PantoMatrix/BEAT2/beat_english_v2.0.0/',
cache_path=
'./cache/beatx_cache_backup/',
debug=False,
tiny=False,
face_joint_idx=[18, 13, 9, 5],
sample_rate=16000,
num_mels=80,
hop_length=512,
fps=15,
stride=5,
pose_length=150,
ori_joints='beat_smplx_joints',
deps_path='/CT/GestureSynth1/work/GestureGPT/GestureRep/deps/',
training_speakers=[
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
],
additional_data=True,
pose_fps=15,
audio_sr=16000,
audio_fps=16000,
new_cache=False,
clean_first_seconds=0,
clean_final_seconds=0,
audio_rep='wav2vec',
word_rep='bert_framealigned',
id_rep='idx',
sem_rep='info',
prom_rep='prom',
emo_rep='emo',
split='test')
data = dict(
samples_per_gpu=128,
workers_per_gpu=8,
train=dict(
type='BEATXDataset',
pose_rep='smplxflame_30',
facial_rep='smplxflame_30',
data_path=
'/CT/GestureSynth1/work/GestureGPT/PantoMatrix/BEAT2/beat_english_v2.0.0/',
cache_path=
'./cache/beatx_cache_backup/',
debug=False,
tiny=False,
face_joint_idx=[18, 13, 9, 5],
sample_rate=16000,
num_mels=80,
hop_length=512,
fps=15,
stride=5,
pose_length=150,
ori_joints='beat_smplx_joints',
deps_path='/CT/GestureSynth1/work/GestureGPT/GestureRep/deps/',
training_speakers=[
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
],
additional_data=True,
pose_fps=15,
audio_sr=16000,
audio_fps=16000,
new_cache=False,
clean_first_seconds=0,
clean_final_seconds=0,
audio_rep='wav2vec',
word_rep='bert_framealigned',
id_rep='idx',
sem_rep='info',
prom_rep='prom',
emo_rep='emo',
split='train'),
val=dict(
type='BEATXDataset',
pose_rep='smplxflame_30',
facial_rep='smplxflame_30',
data_path=
'/CT/GestureSynth1/work/GestureGPT/PantoMatrix/BEAT2/beat_english_v2.0.0/',
cache_path=
'/CT/GestureSynth1/work/DiscourseAwareGesture/RAGGesture_BEATX/cache/beatx_cache_backup/',
debug=False,
tiny=False,
face_joint_idx=[18, 13, 9, 5],
sample_rate=16000,
num_mels=80,
hop_length=512,
fps=15,
stride=5,
pose_length=150,
ori_joints='beat_smplx_joints',
deps_path='/CT/GestureSynth1/work/GestureGPT/GestureRep/deps/',
training_speakers=[
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
],
additional_data=True,
pose_fps=15,
audio_sr=16000,
audio_fps=16000,
new_cache=False,
clean_first_seconds=0,
clean_final_seconds=0,
audio_rep='wav2vec',
word_rep='bert_framealigned',
id_rep='idx',
sem_rep='info',
prom_rep='prom',
emo_rep='emo',
split='val'),
test=dict(
type='BEATXDataset',
pose_rep='smplxflame_30',
facial_rep='smplxflame_30',
data_path=
'/CT/GestureSynth1/work/GestureGPT/PantoMatrix/BEAT2/beat_english_v2.0.0/',
cache_path=
'./cache/beatx_cache_backup/',
debug=False,
tiny=False,
face_joint_idx=[18, 13, 9, 5],
sample_rate=16000,
num_mels=80,
hop_length=512,
fps=15,
stride=5,
pose_length=150,
ori_joints='beat_smplx_joints',
deps_path='/CT/GestureSynth1/work/GestureGPT/GestureRep/deps/',
training_speakers=[
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
],
additional_data=True,
pose_fps=15,
audio_sr=16000,
audio_fps=16000,
new_cache=False,
clean_first_seconds=0,
clean_final_seconds=0,
audio_rep='wav2vec',
word_rep='bert_framealigned',
id_rep='idx',
sem_rep='info',
prom_rep='prom',
emo_rep='emo',
split='test'))
checkpoint_config = dict(interval=2)
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = None
resume_from = './experiments/base_beatx_len150fps15_lowertrans4lats_remodiffuseclfguide_8l16h/latest.pth'
workflow = [('train', 1)]
optimizer = dict(type='Adam', lr=0.0001)
optimizer_config = dict(grad_clip=None)
lr_config = dict(policy='CosineAnnealing', min_lr_ratio=1e-06, by_epoch=False)
runner = dict(type='EpochBasedRunner', max_epochs=500)
log_config = dict(
interval=10,
hooks=[dict(type='TextLoggerHook'),
dict(type='TensorboardLoggerHook')])
custom_hooks = [
dict(type='VAE_FreezeHook'),
dict(
type='DatabaseSaveHook',
save_dir=
'./experiments/retrieval_dicts'
)
]
input_feats = 189
max_seq_len = 150
frame_chunk_size = 15
latent_dim = 512
time_embed_dim = 2048
inp_text_latent_dim = 768
ff_size = 1024
num_heads = 16
num_layers = 8
dropout = 0
model = dict(
type='MotionDiffusion',
model=dict(
type='ReGestureTransformer',
input_feats=189,
max_seq_len=150,
frame_chunk_size=15,
latent_dim=512,
time_embed_dim=2048,
num_layers=8,
body_part_cat_axis='time',
sa_block_cfg=dict(
type='EfficientSelfAttention',
latent_dim=512,
num_heads=16,
dropout=0,
time_embed_dim=2048),
ca_block_cfg=dict(
type='EfficientCrossAttention',
latent_dim=512,
text_latent_dim=512,
num_heads=16,
dropout=0,
time_embed_dim=2048),
ffn_cfg=dict(
latent_dim=512, ffn_dim=1024, dropout=0, time_embed_dim=2048),
vae_cfg=dict(
upper_cfg=
'/CT/GestureSynth1/work/GestureGPT/GestureRep/experiments/0903_020101_gesture_lexicon_transformer_vae_upper_allspk_len256_l8h4_fchunksize15/0903_020101_gesture_lexicon_transformer_vae_upper_allspk_len256_l8h4_fchunksize15.yaml',
lowertrans_cfg=
'/CT/GestureSynth1/work/GestureGPT/GestureRep/experiments/1031_142417_gesture_lexicon_transformer_vae_lowerplustrans_allspk_len10s_l8h8_fchunksize15_run2/1031_142417_gesture_lexicon_transformer_vae_lowerplustrans_allspk_len10s_l8h8_fchunksize15_run2.yaml',
face_cfg=
'/CT/GestureSynth1/work/GestureGPT/GestureRep/experiments/0909_130750_gesture_lexicon_transformer_vae_face_allspk_len10s_l8h4_fchunksize15/0909_130750_gesture_lexicon_transformer_vae_face_allspk_len10s_l8h4_fchunksize15.yaml',
hands_cfg=
'/CT/GestureSynth1/work/GestureGPT/GestureRep/experiments/0909_132647_gesture_lexicon_transformer_vae_hands_allspk_len10s_l8h4_fchunksize15/0909_132647_gesture_lexicon_transformer_vae_hands_allspk_len10s_l8h4_fchunksize15.yaml',
latent_dim=512,
frame_chunk_size=15),
text_encoder=dict(
pretrained_model=None,
latent_dim=768,
num_layers=0,
ff_size=2048,
dropout=0,
use_text_proj=False),
audio_encoder=dict(
pretrained_model=None, latent_dim=768, num_layers=0, dropout=0.1),
speaker_embedding=dict(num_speakers=25),
retrieval_train=False,
retrieval_cfg=dict(
motion_feat_dim=189,
num_retrieval=1,
stride=4,
num_layers=2,
num_motion_layers=2,
kinematic_coef=0.1,
topk=2,
latent_dim=512,
text_latent_dim=768,
output_dim=512,
max_seq_len=150,
motion_fps=15,
motion_framechunksize=15,
num_heads=16,
ff_size=1024,
dropout=0,
lmdb_paths=
'/CT/GestureSynth1/work/DiscourseAwareGesture/RAGGesture_BEATX/cache/retrieval_cache_backup/retrieval_cache_stratified/',
new_lmdb_cache=False,
stratified_db_creation=True,
stratification_interval=15,
ffn_cfg=dict(latent_dim=512, ffn_dim=1024, dropout=0),
sa_block_cfg=dict(
type='EfficientSelfAttention',
latent_dim=512,
num_heads=16,
dropout=0)),
scale_func_cfg=dict(
coarse_scale=6.5,
both_coef=0.52351,
text_coef=-0.28419,
retr_coef=2.39872),
per_joint_scale=dict(upper=1.0, hands=1.0, face=1.0, lowertransl=1.0),
),
loss_recon=dict(type='MSELoss', loss_weight=1, reduction='none'),
body_part_lossweights=dict(
upper=1.0, hands=1.0, face=1.0, lowertransl=1.0),
diffusion_train=dict(
beta_scheduler='scaled_linear',
diffusion_steps=1000,
model_mean_type='start_x',
model_var_type='fixed_large'),
diffusion_test=dict(
beta_scheduler='scaled_linear',
diffusion_steps=1000,
model_mean_type='start_x',
model_var_type='fixed_large',
respace='15,15,8,6,6',
num_inference_timesteps=50,
classifier_free_guidance_scale=0),
inference_type='ddim')
work_dir = './experiments/base_beatx_len150fps15_lowertrans4lats_remodiffuseclfguide_8l16h/'
gpu_ids = range(0, 1)