MotionLCM / configs /motionlcm_t2m_clt.yaml
wxDai's picture
[Init]
c64dfa4
FOLDER: './experiments_t2m'
TEST_FOLDER: './experiments_t2m_test'
NAME: 'motionlcm_humanml'
SEED_VALUE: 1234
TEST:
BATCH_SIZE: 1
SPLIT: 'test'
NUM_WORKERS: 12
PERSISTENT_WORKERS: true
CHECKPOINTS: 'experiments_t2m/motionlcm_humanml/motionlcm_humanml.ckpt'
# Testing Args
REPLICATION_TIMES: 1
DIVERSITY_TIMES: 300
DO_MM_TEST: false
MAX_NUM_SAMPLES: 1024
DATASET:
NAME: 'humanml3d'
SMPL_PATH: './deps/smpl'
WORD_VERTILIZER_PATH: './deps/glove/'
HUMANML3D:
FRAME_RATE: 20.0
UNIT_LEN: 4
ROOT: './datasets/humanml3d'
CONTROL_ARGS:
CONTROL: true
TEMPORAL: false
TRAIN_JOINTS: [0]
TEST_JOINTS: [0]
TRAIN_DENSITY: 'random'
TEST_DENSITY: 100
MEAN_STD_PATH: './datasets/humanml_spatial_norm'
SAMPLER:
MAX_LEN: 200
MIN_LEN: 40
MAX_TEXT_LEN: 20
PADDING_TO_MAX: false
WINDOW_SIZE: null
METRIC:
DIST_SYNC_ON_STEP: true
TYPE: ['TM2TMetrics', 'ControlMetrics']
model:
target: ['motion_vae', 'text_encoder', 'denoiser', 'scheduler_lcm', 'noise_optimizer']
latent_dim: [16, 32]
guidance_scale: 'dynamic'
t2m_textencoder:
dim_word: 300
dim_pos_ohot: 15
dim_text_hidden: 512
dim_coemb_hidden: 512
t2m_motionencoder:
dim_move_hidden: 512
dim_move_latent: 512
dim_motion_hidden: 1024
dim_motion_latent: 512
bert_path: './deps/distilbert-base-uncased'
clip_path: './deps/clip-vit-large-patch14'
t5_path: './deps/sentence-t5-large'
t2m_path: './deps/t2m/'