FOLDER: './experiments_control/temporal'
TEST_FOLDER: './experiments_control_test/temporal'

NAME: 'motionlcm_humanml'

SEED_VALUE: 1234

TRAIN:
  DATASET: 'humanml3d'
  BATCH_SIZE: 128
  SPLIT: 'train'
  NUM_WORKERS: 8
  PERSISTENT_WORKERS: true

  PRETRAINED: 'experiments_t2m/motionlcm_humanml/motionlcm_humanml.ckpt'

  validation_steps: -1
  validation_epochs: 50
  checkpointing_steps: -1
  checkpointing_epochs: 50
  max_train_steps: -1
  max_train_epochs: 1000
  learning_rate: 1e-4
  learning_rate_spatial: 1e-4
  lr_scheduler: "cosine"
  lr_warmup_steps: 1000
  adam_beta1: 0.9
  adam_beta2: 0.999
  adam_weight_decay: 0.0
  adam_epsilon: 1e-08
  max_grad_norm: 1.0

VAL:
  DATASET: 'humanml3d'
  BATCH_SIZE: 32
  SPLIT: 'test'
  NUM_WORKERS: 12
  PERSISTENT_WORKERS: true

TEST:
  DATASET: 'humanml3d'
  BATCH_SIZE: 32
  SPLIT: 'test'
  NUM_WORKERS: 12
  PERSISTENT_WORKERS: true

  CHECKPOINTS: 'experiments_control/temporal/motionlcm_humanml/motionlcm_humanml_t.ckpt'

  # Testing Args
  REPLICATION_TIMES: 20
  DIVERSITY_TIMES: 300
  DO_MM_TEST: false

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: true
      TRAIN_JOINTS: [0, 10, 11, 15, 20, 21]
      TEST_JOINTS: [0, 10, 11, 15, 20, 21]
      TRAIN_DENSITY: [25, 25]
      TEST_DENSITY: 25
      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', 'traj_encoder', 'noise_optimizer']
  latent_dim: [16, 32]
  guidance_scale: 'dynamic'

  # ControlNet Args
  is_controlnet: true
  vaeloss: true
  vaeloss_type: 'sum'
  cond_ratio: 1.0
  control_loss_func: 'l2'
  use_3d: false
  lcm_w_min_nax: [5, 15]
  lcm_num_ddim_timesteps: 10

  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/'