Upload 16 files

checkpoints/epoch=119.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:790f0cababdc401ca0455ddff64ef7ddb7082f5706e30f29e9bd87189c6c4f67
+size 1130308662

checkpoints/epoch=149.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f70cc12964dff35aa733c92d1e1d1f9edec52a1f348f5bd754904b5708b76e96
+size 1130309045

checkpoints/epoch=179.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e781b61aac30c5569790f9c09d43a5bd635f069b85ce199abecbaf27568c6bc
+size 1130309428

checkpoints/epoch=209.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bbe35341bb9434977a93a9dc4c176ee0bd9144b8e0788b272491aeecf774d4f9
+size 1130309811

checkpoints/epoch=239.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdb38ceae6149b94d3c5f171251c6719953dcd4468ce9018d93919fa2ae80d36
+size 1130310194

checkpoints/epoch=269.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:486846f0f378342ba43c68bf0ff95c37a561b479585ccd8446072ec119636015
+size 1130310577

checkpoints/epoch=299.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:091bae9ef1807cd69f3a2a9773b93f2e909e69fea75b6bfb9985987b0ff78736
+size 1130310960

checkpoints/epoch=329.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31d2f4447924b8d3ea89228d028d3f676e741b4a292eb8e1a59c24b210148e74
+size 1130311343

checkpoints/epoch=359.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11c4eab8e78db60b9ef9f8f99e950eefcd8f9b335e36c0686c10b92e078f39fc
+size 1130311790

checkpoints/epoch=389.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b23d0d1aa700ca5e8cca573dbae90aa96da6bd1603536ef3efa6146d2fb69d3
+size 1130312173

checkpoints/epoch=49.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ec566353eec79dd9fa60cf721177acf2251840d9591726d329dd5039d2350ed
+size 1130308662

checkpoints/epoch=99.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cdf355a4970e4ab3979a238dfc37dcbf04dfef8c1fd2c20a2c860b312c0f398c
+size 1130309045

configs/data/assets.yaml

@@ -0,0 +1,18 @@
+FOLDER: './checkpoints' # Experiment files saving path
+
+TEST:
+  FOLDER: './outputs' # Testing files saving path
+
+DATASET:
+  SMPL_PATH: '/data2/lrh/smpl_model/smpl'
+  SMPLH_PATH: '/data2/lrh/smpl_model/smplh'
+  SMPLX_PATH: '/data2/lrh/smpl_model/smplx'
+  FINEDANCE_139CUT:
+    ROOT: data/finedance
+    MOTION: data/finedance/mofea319
+    MUSIC: data/finedance/music_npynew
+    normalizer: data/Normalizer.pth
+    LABEL: data/finedance/label_json/
+
+  
+

configs/lodge/coarse_finedance_fea139.yaml

@@ -0,0 +1,158 @@
+NAME: FineDance_Coarse_v2_relative_noNorm_128len_139_diff_bc800 # Experiment name
+ACCELERATOR: 'gpu' # Devices optioncal: “cpu”, “gpu”, “tpu”, “ipu”, “hpu”, “mps, “auto”
+DEVICE: [0,1,2,3] 
+
+FOLDER: ./checkpoints
+SEED_VALUE: 1234
+Norm: True
+DEBUG: False
+Discriminator: False
+FineTune: False
+TRAIN:
+  STAGE: diffusion    # diffusion
+  SPLIT: 'train'
+  NUM_WORKERS: 16 # Number of workers
+  BATCH_SIZE: 190 # Size of batches
+  START_EPOCH: 1 # Start epoch
+  END_EPOCH: 3001 # End epoch
+  RESUME: '/home/seungho/research/lodge/experiments/Global_Module/debug--FineDance_Coarse_v2_relative_noNorm_128len_139_diff_bc800/' # Experiment path to be resumed training
+  PRETRAINED_VAE: ''
+  PRETRAINED: 'checkpoints/exp/Local_Module/FineDance_FineTuneV2_Local/checkpoints/epoch=299.ckpt' # Attention! Set your Pretrained model path
+  DATASETS: ['FINEDANCE_139CUT'] # Training datasets
+
+  # OPTIM:
+  #   TYPE: 'AdamW' # Optimizer type
+  #   LR: 1e-4 # Learning rate
+  OPTIM:
+    TYPE: 'Adan' # Optimizer type
+    LR: 4e-4 # Learning rate
+
+  ABLATION:
+    VAE_TYPE: 'no' # vae ablation: actor or mcross
+    VAE_ARCH: 'encoder_decoder' # mdiffusion vae architecture
+    PE_TYPE: 'actor' # mdiffusion mld or actor
+    DIFF_PE_TYPE: 'actor' # mdiffusion mld or actor
+    SKIP_CONNECT: False # skip connection for denoiser va
+    # use linear to expand mean and std rather expand token nums
+    MLP_DIST: False
+    IS_DIST: False # Mcross distribution kl
+    PREDICT_EPSILON: False # noise or motion
+
+EVAL:
+  SPLIT: 'gtest'
+  BATCH_SIZE: 32 # Evaluating Batch size
+  NUM_WORKERS: 16 # Evaluating Batch size
+
+TEST:
+  TEST_DIR: ''
+  CHECKPOINTS: ''
+  SPLIT: 'gtest'
+  BATCH_SIZE: 32 # Testing Batch size
+  NUM_WORKERS: 12 # Evaluating Batch size
+  SAVE_PREDICTIONS: True # Weather to save predictions
+  COUNT_TIME: False # Weather to count time during test
+  REPLICATION_TIMES: 2 # Number of times to replicate the test
+  MM_NUM_SAMPLES: 100 # Number of samples for multimodal test
+  MM_NUM_REPEATS: 30 # Number of repeats for multimodal test
+  MM_NUM_TIMES: 10 # Number of times to repeat the multimodal test
+  DIVERSITY_TIMES: 300 # Number of times to repeat the diversity test
+  REP_I: 0
+  DATASETS: ['FINEDANCE_139CUT']
+
+model:
+  target: 'modules'
+  vae_type: 'no'
+  
+  vae: true # whether vae model
+  model_type: Global_Module # model type
+  condition: 'music'
+  latent_dim: [1, 256]  #[512, 135]    # [${FINEDANCE.full_seq_len}, ${DATASET.NFEATS}]      #  # latent dimension
+  ff_size: 1024 #
+  num_layers: 9 # number of layers
+  num_head: 4 # number of head layers
+  droupout: 0.1 # dropout rate
+  activation: gelu # activation type
+  guidance_scale: 7.5   # 7.5 #
+  guidance_uncondp: 0.1 # 0.1 0.25
+
+  diffusion:   
+    target: dld.models.architectures.diffusion.GaussianDiffusion
+    params:
+      horizon: ${FINEDANCE.full_seq_len}
+      repr_dim: ${FINEDANCE.nfeats}
+      schedule: "cosine"
+      n_timestep: 1000
+      predict_epsilon: False
+      loss_type: "l2"
+      use_p2: False
+      cond_drop_prob: 0.25
+      guidance_weight: 2
+ 
+  DanceDecoder:   
+    target: dld.models.architectures.model.DanceDecoderv2
+    params:
+      nfeats: ${FINEDANCE.nfeats}
+      seq_len: ${FINEDANCE.full_seq_len}
+      latent_dim: 512
+      ff_size: 1024
+      num_layers: 8
+      num_heads: 8
+      dropout: 0.1
+      cond_feature_dim: 35
+      activation: gelu
+        
+
+LOSS:
+  TYPE: smpl_loss # Losses type
+  LAMBDA_MSE: 0.636 
+  LAMBDA_V: 2.964 
+  LAMBDA_FK: 0.636
+  LAMBDA_FOOT: 10.942 # Lambda for reconstruction losses
+  DIST_SYNC_ON_STEP: True # Sync Losses on step when distributed trained
+METRIC:
+  FORCE_IN_METER: True
+  DIST_SYNC_ON_STEP: True  # # Sync Losses on step when distributed trained
+  TYPE: ['DanceDiffuse_Metric']  # DanceAE_Metric
+LOGGER:
+  SACE_CHECKPOINT_EPOCH: 30      
+  LOG_EVERY_STEPS: 30
+  VAL_EVERY_STEPS: 30
+  TENSORBOARD: true
+  WANDB:
+    OFFLINE: false
+    PROJECT: null
+    RESUME_ID: null
+RENDER:
+  JOINT_TYPE: mmm
+  INPUT_MODE: npy
+  DIR: ''
+  NPY: ''
+  DENOISING: true
+  OLDRENDER: true
+  RES: high
+  DOWNSAMPLE: true
+  FPS: 12.5
+  CANONICALIZE: true
+  EXACT_FRAME: 0.5
+  NUM: 7
+  MODE: sequence
+  VID_EXT: mp4
+  ALWAYS_ON_FLOOR: false
+  GT: false
+DEMO:
+  MusicDir: /data2/lrh/dataset/fine_dance/origin/music
+  RENDER: false
+  EXAMPLE: null
+  use_cached_features: false
+
+FINEDANCE:
+  mix: False           # 是否做混合的数据增强
+  full_seq_len: 256   # 150
+  length_fi: 128
+  windows: 10  # 16   #40   #
+  is_mirror: False    # 是否做镜像的数据增强
+  nfeats: 139
+  njoints: 22   
+  mode: single  # "single" , "double_react", "double"
+  partial: full   # "full" , "morden", "tradition"
+  GENRE_NUM: 16

configs/lodge/finedance_fea139.yaml

@@ -0,0 +1,170 @@
+NAME: FineDance_relative_Norm_GenreDis_bc190 # Experiment name
+ACCELERATOR: 'gpu' # Devices optioncal: “cpu”, “gpu”, “tpu”, “ipu”, “hpu”, “mps, “auto”
+DEVICE: [0]   #,  #  # Index of gpus eg. [0] or [0,1,2,3]
+
+# FOLDER: ./experiments
+SEED_VALUE: 7777
+Norm: True
+Discriminator: True
+DEBUG: False
+FineTune: false
+TRAIN:
+  STAGE: diffusion    # diffusion
+  SPLIT: 'train'
+  NUM_WORKERS: 16 # Number of workers
+  BATCH_SIZE: 64 # Size of batches
+  START_EPOCH: 1 # Start epoch
+  END_EPOCH: 3001 # End epoch
+  RESUME: '' # Experiment path to be resumed training
+  # PRETRAINED_VAE: ''
+  PRETRAINED: '' # Pretrained model path
+  DATASETS: ['FINEDANCE_139CUT'] # Training datasets
+
+  OPTIM:
+    TYPE: 'Adan' # Optimizer type
+    LR: 4e-4 # Learning rate
+
+  ABLATION:
+    VAE_TYPE: 'no' # vae ablation: actor or mcross
+    VAE_ARCH: 'encoder_decoder' # mdiffusion vae architecture
+    PE_TYPE: 'actor' # mdiffusion mld or actor
+    DIFF_PE_TYPE: 'actor' # mdiffusion mld or actor
+    SKIP_CONNECT: False # skip connection for denoiser va
+    # use linear to expand mean and std rather expand token nums
+    MLP_DIST: False
+    IS_DIST: False # Mcross distribution kl
+    PREDICT_EPSILON: False # noise or motion
+
+EVAL:
+  SPLIT: 'gtest'
+  BATCH_SIZE: 32 # Evaluating Batch size
+  NUM_WORKERS: 16 # Evaluating Batch size
+
+TEST:
+  TEST_DIR: ''
+  CHECKPOINTS: '' #/data2/lrh/project/dance/Lodge/lodge_pylight/experiments/Edge_Module/0115_newNorm_256len_263_diff_bc768/checkpoints/epoch=3899.ckpt 
+  SPLIT: 'gtest'
+  BATCH_SIZE: 32 # Testing Batch size
+  NUM_WORKERS: 12 # Evaluating Batch size
+  SAVE_PREDICTIONS: True # Weather to save predictions
+  COUNT_TIME: False # Weather to count time during test
+  REPLICATION_TIMES: 2 # Number of times to replicate the test
+  MM_NUM_SAMPLES: 100 # Number of samples for multimodal test
+  MM_NUM_REPEATS: 30 # Number of repeats for multimodal test
+  MM_NUM_TIMES: 10 # Number of times to repeat the multimodal test
+  DIVERSITY_TIMES: 300 # Number of times to repeat the diversity test
+  REP_I: 0
+  DATASETS: ['FINEDANCE_139CUT']
+
+model:
+  target: 'modules'
+  vae_type: 'no'
+  
+  vae: true # whether vae model
+  model_type: Local_Module # model type
+  condition: 'music'
+  latent_dim: [1, 256]  #[512, 135]    # [${FINEDANCE.full_seq_len}, ${DATASET.NFEATS}]      #  # latent dimension
+  ff_size: 1024 #
+  num_layers: 9 # number of layers
+  num_head: 4 # number of head layers
+  droupout: 0.1 # dropout rate
+  activation: gelu # activation type
+  guidance_scale: 7.5   # 7.5 #
+  guidance_uncondp: 0.1 # 0.1 0.25
+
+
+  diffusion:   
+    target: dld.models.architectures.diffusion.GaussianDiffusion
+    params:
+      horizon: ${FINEDANCE.full_seq_len}
+      repr_dim: ${FINEDANCE.nfeats}
+      schedule: "cosine"
+      n_timestep: 1000
+      predict_epsilon: False
+      loss_type: "l2"
+      use_p2: False
+      cond_drop_prob: 0.25
+      guidance_weight: 2
+ 
+  DanceDecoder:   
+    target: dld.models.architectures.model.Refine_DanceDecoder
+    params:
+      nfeats: ${FINEDANCE.nfeats}
+      seq_len: ${FINEDANCE.full_seq_len}
+      latent_dim: 512
+      ff_size: 1024
+      num_layers: 8
+      num_heads: 8
+      dropout: 0.1
+      cond_feature_dim: 35
+      activation: gelu
+
+  DanceDiscriminator:   
+    target: dld.models.architectures.model.DanceDiscriminator
+    params:
+      nfeats: ${FINEDANCE.nfeats}
+      seq_len: ${FINEDANCE.full_seq_len}
+      latent_dim: 512
+      ff_size: 1024
+      num_layers: 8
+      num_heads: 8
+      dropout: 0.1
+      cond_feature_dim: 35
+      activation: gelu
+        
+LOSS:
+  TYPE: smpl_loss_relative # Losses type
+  LAMBDA_MSE: 0.636 
+  LAMBDA_V: 2.964 
+  LAMBDA_FK: 1 
+  LAMBDA_FOOT: 20 # Lambda for reconstruction losses
+  LAMBDA_FC: 0
+  LAMBDA_TRANS: 0
+  LAMBDA_DIS: 0.1
+  DIST_SYNC_ON_STEP: True # Sync Losses on step when distributed trained
+METRIC:
+  FORCE_IN_METER: True
+  DIST_SYNC_ON_STEP: True  # # Sync Losses on step when distributed trained
+  TYPE: ['DanceDiffuse_Metric']  # DanceAE_Metric
+LOGGER:
+  SACE_CHECKPOINT_EPOCH: 100      # 保存间隔
+  LOG_EVERY_STEPS: 100
+  VAL_EVERY_STEPS: 100
+  TENSORBOARD: true
+  WANDB:
+    OFFLINE: false
+    PROJECT: null
+    RESUME_ID: null
+RENDER:
+  JOINT_TYPE: mmm
+  INPUT_MODE: npy
+  DIR: ''
+  NPY: ''
+  DENOISING: true
+  OLDRENDER: true
+  RES: high
+  DOWNSAMPLE: true
+  FPS: 12.5
+  CANONICALIZE: true
+  EXACT_FRAME: 0.5
+  NUM: 7
+  MODE: sequence
+  VID_EXT: mp4
+  ALWAYS_ON_FLOOR: false
+  GT: false
+DEMO:
+  MusicDir: /data2/lrh/dataset/fine_dance/origin/music
+  RENDER: false
+  EXAMPLE: null
+  use_cached_features: false
+
+FINEDANCE:
+  mix: False           # weather do mix data augmention
+  full_seq_len: 256   # 150
+  windows: 8  # 16   #40   #
+  is_mirror: False    # weather do mirror data augmention
+  nfeats: 139
+  njoints: 22   
+  mode: single  # "single" , "double_react", "double"
+  partial: full   # "full" , "morden", "tradition"
+  GENRE_NUM: 16

configs/lodge/finedance_fea139_finetune_v2.yaml

@@ -0,0 +1,169 @@
+NAME: Debug_FineDance_FineTuneV2_Local # Experiment name
+ACCELERATOR: 'gpu' # Devices optioncal: “cpu”, “gpu”, “tpu”, “ipu”, “hpu”, “mps, “auto”
+DEVICE: [0,1,2,3]   #,  #  # Index of gpus eg. [0] or [0,1,2,3]
+
+# FOLDER: ./experiments
+SEED_VALUE: 1233
+Norm: True
+Discriminator: True
+DEBUG: False
+FineTune: True
+TRAIN:
+  STAGE: diffusion    # diffusion
+  SPLIT: 'train'
+  NUM_WORKERS: 16 # Number of workers
+  BATCH_SIZE: 190 # Size of batches
+  START_EPOCH: 1 # Start epoch
+  END_EPOCH: 1601 # End epoch
+  RESUME: '' # Experiment path to be resumed training
+  PRETRAINED: '/data2/lrh/project/dance/Lodge/lodge302/experiments/Local_Module/FineDance_relative_Norm_GenreDis_bc190/checkpoints/epoch=1399.ckpt' # Pretrained model path
+  DATASETS: ['FINEDANCE_139CUT'] # Training datasets
+
+  OPTIM:
+    TYPE: 'Adan' # Optimizer type
+    LR: 4e-4 # Learning rate
+
+  ABLATION:
+    VAE_TYPE: 'no' # vae ablation: actor or mcross
+    VAE_ARCH: 'encoder_decoder' # mdiffusion vae architecture
+    PE_TYPE: 'actor' # mdiffusion mld or actor
+    DIFF_PE_TYPE: 'actor' # mdiffusion mld or actor
+    SKIP_CONNECT: False # skip connection for denoiser va
+    # use linear to expand mean and std rather expand token nums
+    MLP_DIST: False
+    IS_DIST: False # Mcross distribution kl
+    PREDICT_EPSILON: False # noise or motion
+
+EVAL:
+  SPLIT: 'gtest'
+  BATCH_SIZE: 256 # Evaluating Batch size
+  NUM_WORKERS: 16 # Evaluating Batch size
+TEST:
+  TEST_DIR: ''
+  CHECKPOINTS: ''  
+  SPLIT: 'gtest'
+  BATCH_SIZE: 256 # Testing Batch size
+  NUM_WORKERS: 12 # Evaluating Batch size
+  SAVE_PREDICTIONS: True # Weather to save predictions
+  COUNT_TIME: False # Weather to count time during test
+  REPLICATION_TIMES: 2 # Number of times to replicate the test
+  MM_NUM_SAMPLES: 100 # Number of samples for multimodal test
+  MM_NUM_REPEATS: 30 # Number of repeats for multimodal test
+  MM_NUM_TIMES: 10 # Number of times to repeat the multimodal test
+  DIVERSITY_TIMES: 300 # Number of times to repeat the diversity test
+  REP_I: 0
+  DATASETS: ['FINEDANCE_139CUT']
+
+model:
+  target: 'modules'
+  vae_type: 'no'
+  
+  vae: true # whether vae model
+  model_type: Local_Module # model type
+  condition: 'music'
+  latent_dim: [1, 256]  #[512, 135]    # [${FINEDANCE.full_seq_len}, ${DATASET.NFEATS}]      #  # latent dimension
+  ff_size: 1024 #
+  num_layers: 9 # number of layers
+  num_head: 4 # number of head layers
+  droupout: 0.1 # dropout rate
+  activation: gelu # activation type
+  guidance_scale: 7.5   # 7.5 #
+  guidance_uncondp: 0.1 # 0.1 0.25
+
+  diffusion:   
+    target: dld.models.architectures.diffusion.GaussianDiffusion
+    params:
+      horizon: ${FINEDANCE.full_seq_len}
+      repr_dim: ${FINEDANCE.nfeats}
+      schedule: "cosine"
+      n_timestep: 1000
+      predict_epsilon: False
+      loss_type: "l2"
+      use_p2: False
+      cond_drop_prob: 0.25
+      guidance_weight: 2
+ 
+  DanceDecoder:   
+    target: dld.models.architectures.model.Refine_DanceDecoder
+    params:
+      nfeats: ${FINEDANCE.nfeats}
+      seq_len: ${FINEDANCE.full_seq_len}
+      latent_dim: 512
+      ff_size: 1024
+      num_layers: 8
+      num_heads: 8
+      dropout: 0.1
+      cond_feature_dim: 35
+      activation: gelu
+
+  DanceDiscriminator:   
+    target: dld.models.architectures.model.DanceDiscriminator
+    params:
+      nfeats: ${FINEDANCE.nfeats}
+      seq_len: ${FINEDANCE.full_seq_len}
+      latent_dim: 512
+      ff_size: 1024
+      num_layers: 8
+      num_heads: 8
+      dropout: 0.1
+      cond_feature_dim: 35
+      activation: gelu
+        
+LOSS:
+  TYPE: smpl_loss_relative # Losses type
+  LAMBDA_MSE: 0.636
+  LAMBDA_V: 2.964
+  LAMBDA_FK: 1
+  LAMBDA_FK_V: 0
+  LAMBDA_FK_A: 0
+  LAMBDA_FOOT: 10 # Lambda for reconstruction losses
+  LAMBDA_FC: 0
+  LAMBDA_TRANS: 0.5
+  LAMBDA_DIS: 0.1
+  DIST_SYNC_ON_STEP: True # Sync Losses on step when distributed trained
+METRIC:
+  FORCE_IN_METER: True
+  DIST_SYNC_ON_STEP: True  # # Sync Losses on step when distributed trained
+  TYPE: ['DanceDiffuse_Metric']  # DanceAE_Metric
+LOGGER:
+  SACE_CHECKPOINT_EPOCH: 50      # 保存间隔
+  LOG_EVERY_STEPS: 50
+  VAL_EVERY_STEPS: 50
+  TENSORBOARD: true
+  WANDB:
+    OFFLINE: false
+    PROJECT: null
+    RESUME_ID: null
+RENDER:
+  JOINT_TYPE: mmm
+  INPUT_MODE: npy
+  DIR: ''
+  NPY: ''
+  DENOISING: true
+  OLDRENDER: true
+  RES: high
+  DOWNSAMPLE: true
+  FPS: 12.5
+  CANONICALIZE: true
+  EXACT_FRAME: 0.5
+  NUM: 7
+  MODE: sequence
+  VID_EXT: mp4
+  ALWAYS_ON_FLOOR: false
+  GT: false
+DEMO:
+  MusicDir: /data2/lrh/dataset/fine_dance/origin/music
+  RENDER: false
+  EXAMPLE: null
+  use_cached_features: false
+
+FINEDANCE:
+  mix: False           # 是否做混合的数据增强
+  full_seq_len: 256   # 150
+  windows: 8  # 16   #40   #
+  is_mirror: False    # 是否做镜像的数据增强
+  nfeats: 139
+  njoints: 22   
+  mode: single  # "single" , "double_react", "double"
+  partial: full   # "full" , "morden", "tradition"
+  GENRE_NUM: 16