add paper models

contrastive-vc/config.yaml

@@ -0,0 +1,265 @@
+# pytorch_lightning==1.9.3
+seed_everything: 123
+trainer:
+  logger:
+    class_path: pytorch_lightning.loggers.TensorBoardLogger
+    init_args:
+      save_dir: logs
+      name: exp_contrastive_reg_sameclip
+      version: null
+      log_graph: false
+      default_hp_metric: true
+      prefix: ''
+      sub_dir: null
+  enable_checkpointing: true
+  callbacks:
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-val-loss-{epoch}-{step}
+      monitor: loss/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-eer-val-{epoch}-{step}
+      monitor: EER evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-rank-val-{epoch}-{step}
+      monitor: Order evaluation mean proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-alignment-val-{epoch}-{step}
+      monitor: Alignment evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-uniformity-val-{epoch}-{step}
+      monitor: Uniformity evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: cptk-{epoch}-{step}
+      monitor: null
+      verbose: false
+      save_last: null
+      save_top_k: -1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: 25
+      save_on_train_epoch_end: null
+  - class_path: callbacks.evaluation.OrderEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  - class_path: callbacks.evaluation.EEREvaluation
+    init_args:
+      use_more_neg: false
+      log_n_epochs: 5
+      on_train: false
+      use_projection: true
+  - class_path: callbacks.evaluation.HypersphereEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  default_root_dir: null
+  gradient_clip_val: null
+  gradient_clip_algorithm: null
+  num_nodes: 1
+  num_processes: null
+  devices: 1
+  gpus: null
+  auto_select_gpus: null
+  tpu_cores: null
+  ipus: null
+  enable_progress_bar: true
+  overfit_batches: 0.0
+  track_grad_norm: -1
+  check_val_every_n_epoch: 1
+  fast_dev_run: false
+  accumulate_grad_batches: null
+  max_epochs: 100000
+  min_epochs: null
+  max_steps: 1000000000
+  min_steps: null
+  max_time: null
+  limit_train_batches: null
+  limit_val_batches: null
+  limit_test_batches: null
+  limit_predict_batches: null
+  val_check_interval: null
+  log_every_n_steps: 50
+  accelerator: gpu
+  strategy: ddp
+  sync_batchnorm: false
+  precision: 32
+  enable_model_summary: true
+  num_sanity_val_steps: 2
+  resume_from_checkpoint: null
+  profiler: null
+  benchmark: null
+  deterministic: null
+  reload_dataloaders_every_n_epochs: 0
+  auto_lr_find: false
+  replace_sampler_ddp: true
+  detect_anomaly: false
+  auto_scale_batch_size: false
+  plugins: null
+  amp_backend: null
+  amp_level: null
+  move_metrics_to_cpu: false
+  multiple_trainloader_mode: max_size_cycle
+  inference_mode: true
+ckpt_path: null
+model:
+  class_path: models.trainer.ContrastiveTrainer
+  init_args:
+    feature_extractor:
+      spec_layer: melspectogram
+      n_fft: 2048
+      hop_length: 512
+    backbone:
+      backbone: efficientnet_b0
+      pretrained: true
+      embedding_dim: 1000
+    projection:
+      input_dim: 1000
+      output_dim: 128
+      l2_normalize: true
+    optimizer1_init:
+      class_path: torch.optim.Adam
+      init_args:
+        lr: 0.0001
+        weight_decay: 1.0e-05
+    use_contrastive_loss: true
+    temp: 0.2
+    nr_negative: 250
+    decouple: true
+    use_norm_reg: false
+    max_norm_hinge: 4.0
+    norm_hinge_fact: 10.0
+    use_invariance_loss: false
+    fact_inv_loss: 1.0
+    use_covariance_reg: true
+    fact_cov: 100.0
+    use_variance_reg: true
+    fact_var: 25.0
+    gamma: 1.0
+    use_vicreg_loss: false
+    use_align_loss: false
+    fact_align_loss: 0.25
+    fact_unif_loss: 0.5
+    use_uniform_loss: false
+    mask_batch: false
+    compute_test_loss: false
+data:
+  class_path: data.vocals.VocalsDataModule
+  init_args:
+    augs_neg:
+      enable: false
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth: 0
+      pitch_range_parselmouth: 0
+      pitch_shift_parselmouth_prob: 0
+    positive_examples: same_clip
+    dataset_dirs:
+    - tencys_vocals
+    - ghero_vocals_3
+    - ghero_vocals_4
+    batch_size: 120
+    batch_size_val: 120
+    nr_samples: 176000
+    normalize: true
+    num_workers: 40
+    sr: 44100
+    batch_sampling_mode: sample_clips
+    eval_frac: 0.105
+    group_name_is_folder: true
+    group_by_artist: true
+    augs:
+      enable: true
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0.5
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth:
+      - 1
+      - 1.3
+      pitch_range_parselmouth: 1.5
+      pitch_shift_parselmouth_prob: 0.5
+    transform_override: false
+    verbose: true
+    use_random_loader: false
+    max_groups: -1
+    multi_epoch: 1
+    classification: false

contrastive-vc/encoder.yaml

@@ -0,0 +1,3 @@
+backbone: efficientnet_b0
+embedding_dim: 1000
+pretrained: true

contrastive-vc/feature_extractor.yaml

@@ -0,0 +1,3 @@
+hop_length: 512
+n_fft: 2048
+spec_layer: melspectogram

contrastive-vc/hyperparams.yaml

@@ -0,0 +1,8 @@
+encoder:
+  backbone: efficientnet_b0
+  embedding_dim: 1000
+  pretrained: true
+feature_extractor:
+  hop_length: 512
+  n_fft: 2048
+  spec_layer: melspectogram

contrastive-vc/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9f72bb45396412b75b9a0ff811f9ce48c577e524f85f8f0c8fd7a4c57c495fc
+size 38765709

contrastive-vc/model.ts

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7063f1ffff6855495636ba9d863ca932b711dc0bb83a7f5691e62a596dd18c50
+size 38969566

contrastive/config.yaml

@@ -0,0 +1,265 @@
+# pytorch_lightning==1.9.3
+seed_everything: 123
+trainer:
+  logger:
+    class_path: pytorch_lightning.loggers.TensorBoardLogger
+    init_args:
+      save_dir: logs
+      name: exp_contrastive_sameclip_v2
+      version: null
+      log_graph: false
+      default_hp_metric: true
+      prefix: ''
+      sub_dir: null
+  enable_checkpointing: true
+  callbacks:
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-val-loss-{epoch}-{step}
+      monitor: loss/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-eer-val-{epoch}-{step}
+      monitor: EER evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-rank-val-{epoch}-{step}
+      monitor: Order evaluation mean proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-alignment-val-{epoch}-{step}
+      monitor: Alignment evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-uniformity-val-{epoch}-{step}
+      monitor: Uniformity evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: cptk-{epoch}-{step}
+      monitor: null
+      verbose: false
+      save_last: null
+      save_top_k: -1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: 25
+      save_on_train_epoch_end: null
+  - class_path: callbacks.evaluation.OrderEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  - class_path: callbacks.evaluation.EEREvaluation
+    init_args:
+      use_more_neg: false
+      log_n_epochs: 5
+      on_train: false
+      use_projection: true
+  - class_path: callbacks.evaluation.HypersphereEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  default_root_dir: null
+  gradient_clip_val: null
+  gradient_clip_algorithm: null
+  num_nodes: 1
+  num_processes: null
+  devices: null
+  gpus: 1
+  auto_select_gpus: null
+  tpu_cores: null
+  ipus: null
+  enable_progress_bar: true
+  overfit_batches: 0.0
+  track_grad_norm: -1
+  check_val_every_n_epoch: 1
+  fast_dev_run: false
+  accumulate_grad_batches: null
+  max_epochs: 100000
+  min_epochs: null
+  max_steps: 1000000000
+  min_steps: null
+  max_time: null
+  limit_train_batches: null
+  limit_val_batches: null
+  limit_test_batches: null
+  limit_predict_batches: null
+  val_check_interval: null
+  log_every_n_steps: 50
+  accelerator: gpu
+  strategy: null
+  sync_batchnorm: false
+  precision: 32
+  enable_model_summary: true
+  num_sanity_val_steps: 2
+  resume_from_checkpoint: null
+  profiler: null
+  benchmark: null
+  deterministic: null
+  reload_dataloaders_every_n_epochs: 0
+  auto_lr_find: false
+  replace_sampler_ddp: true
+  detect_anomaly: false
+  auto_scale_batch_size: false
+  plugins: null
+  amp_backend: null
+  amp_level: null
+  move_metrics_to_cpu: false
+  multiple_trainloader_mode: max_size_cycle
+  inference_mode: true
+ckpt_path: null
+model:
+  class_path: models.trainer.ContrastiveTrainer
+  init_args:
+    feature_extractor:
+      spec_layer: melspectogram
+      n_fft: 2048
+      hop_length: 512
+    backbone:
+      backbone: efficientnet_b0
+      pretrained: true
+      embedding_dim: 1000
+    projection:
+      input_dim: 1000
+      output_dim: 128
+      l2_normalize: true
+    optimizer1_init:
+      class_path: torch.optim.Adam
+      init_args:
+        lr: 0.0001
+        weight_decay: 1.0e-05
+    use_contrastive_loss: true
+    temp: 0.2
+    nr_negative: 250
+    decouple: true
+    use_norm_reg: false
+    max_norm_hinge: 4.0
+    norm_hinge_fact: 10.0
+    use_invariance_loss: false
+    fact_inv_loss: 1.0
+    use_covariance_reg: false
+    fact_cov: 1.0
+    use_variance_reg: false
+    fact_var: 1.0
+    gamma: 1.0
+    use_vicreg_loss: false
+    use_align_loss: false
+    fact_align_loss: 0.25
+    fact_unif_loss: 0.5
+    use_uniform_loss: false
+    mask_batch: false
+    compute_test_loss: false
+data:
+  class_path: data.vocals.VocalsDataModule
+  init_args:
+    augs_neg:
+      enable: false
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth: 0
+      pitch_range_parselmouth: 0
+      pitch_shift_parselmouth_prob: 0
+    positive_examples: same_clip
+    dataset_dirs:
+    - tencys_vocals
+    - ghero_vocals_3
+    - ghero_vocals_4
+    batch_size: 140
+    batch_size_val: 120
+    nr_samples: 176000
+    normalize: true
+    num_workers: 40
+    sr: 44100
+    batch_sampling_mode: sample_clips
+    eval_frac: 0.1
+    group_name_is_folder: true
+    group_by_artist: true
+    augs:
+      enable: true
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0.5
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth:
+      - 1
+      - 1.3
+      pitch_range_parselmouth: 1.5
+      pitch_shift_parselmouth_prob: 0.5
+    transform_override: false
+    verbose: true
+    use_random_loader: false
+    max_groups: -1
+    multi_epoch: 1
+    classification: false

contrastive/encoder.yaml

@@ -0,0 +1,3 @@
+backbone: efficientnet_b0
+embedding_dim: 1000
+pretrained: true

contrastive/feature_extractor.yaml

@@ -0,0 +1,3 @@
+hop_length: 512
+n_fft: 2048
+spec_layer: melspectogram

contrastive/hyperparams.yaml

@@ -0,0 +1,8 @@
+encoder:
+  backbone: efficientnet_b0
+  embedding_dim: 1000
+  pretrained: false
+feature_extractor:
+  hop_length: 512
+  n_fft: 2048
+  spec_layer: melspectogram

contrastive/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:381f2964da906ed85ddaa2453f2ce7a16b84d741ad0f23b1e5357972f2c2692b
+size 38786209

contrastive/model.ts

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97f17b9de1848877ed16801cb873b9c615fcf65cc23fa1c43dfa0d0fe948f40a
+size 38995676

unif-align/config.yaml

@@ -0,0 +1,265 @@
+# pytorch_lightning==1.9.3
+seed_everything: 123
+trainer:
+  logger:
+    class_path: pytorch_lightning.loggers.TensorBoardLogger
+    init_args:
+      save_dir: logs
+      name: exp_uniformity_sameclip
+      version: null
+      log_graph: false
+      default_hp_metric: true
+      prefix: ''
+      sub_dir: null
+  enable_checkpointing: true
+  callbacks:
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-val-loss-{epoch}-{step}
+      monitor: loss/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-eer-val-{epoch}-{step}
+      monitor: EER evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-rank-val-{epoch}-{step}
+      monitor: Order evaluation mean proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-alignment-val-{epoch}-{step}
+      monitor: Alignment evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-uniformity-val-{epoch}-{step}
+      monitor: Uniformity evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: cptk-{epoch}-{step}
+      monitor: null
+      verbose: false
+      save_last: null
+      save_top_k: -1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: 25
+      save_on_train_epoch_end: null
+  - class_path: callbacks.evaluation.OrderEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  - class_path: callbacks.evaluation.EEREvaluation
+    init_args:
+      use_more_neg: false
+      log_n_epochs: 5
+      on_train: false
+      use_projection: true
+  - class_path: callbacks.evaluation.HypersphereEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  default_root_dir: null
+  gradient_clip_val: null
+  gradient_clip_algorithm: null
+  num_nodes: 1
+  num_processes: null
+  devices: null
+  gpus: 2
+  auto_select_gpus: null
+  tpu_cores: null
+  ipus: null
+  enable_progress_bar: true
+  overfit_batches: 0.0
+  track_grad_norm: -1
+  check_val_every_n_epoch: 1
+  fast_dev_run: false
+  accumulate_grad_batches: null
+  max_epochs: 100000
+  min_epochs: null
+  max_steps: 1000000000
+  min_steps: null
+  max_time: null
+  limit_train_batches: null
+  limit_val_batches: null
+  limit_test_batches: null
+  limit_predict_batches: null
+  val_check_interval: null
+  log_every_n_steps: 50
+  accelerator: gpu
+  strategy: null
+  sync_batchnorm: false
+  precision: 32
+  enable_model_summary: true
+  num_sanity_val_steps: 2
+  resume_from_checkpoint: null
+  profiler: null
+  benchmark: null
+  deterministic: null
+  reload_dataloaders_every_n_epochs: 0
+  auto_lr_find: false
+  replace_sampler_ddp: true
+  detect_anomaly: false
+  auto_scale_batch_size: false
+  plugins: null
+  amp_backend: null
+  amp_level: null
+  move_metrics_to_cpu: false
+  multiple_trainloader_mode: max_size_cycle
+  inference_mode: true
+ckpt_path: null
+model:
+  class_path: models.trainer.ContrastiveTrainer
+  init_args:
+    feature_extractor:
+      spec_layer: melspectogram
+      n_fft: 2048
+      hop_length: 512
+    backbone:
+      backbone: efficientnet_b0
+      pretrained: true
+      embedding_dim: 1000
+    projection:
+      input_dim: 1000
+      output_dim: 128
+      l2_normalize: true
+    optimizer1_init:
+      class_path: torch.optim.Adam
+      init_args:
+        lr: 0.0001
+        weight_decay: 1.0e-05
+    use_contrastive_loss: false
+    temp: 0.1
+    nr_negative: 64
+    decouple: false
+    use_norm_reg: false
+    max_norm_hinge: 4.0
+    norm_hinge_fact: 10.0
+    use_invariance_loss: false
+    fact_inv_loss: 1.0
+    use_covariance_reg: false
+    fact_cov: 1.0
+    use_variance_reg: false
+    fact_var: 1.0
+    gamma: 1.0
+    use_vicreg_loss: false
+    use_align_loss: true
+    fact_align_loss: 1.0
+    fact_unif_loss: 1.0
+    use_uniform_loss: true
+    mask_batch: false
+    compute_test_loss: false
+data:
+  class_path: data.vocals.VocalsDataModule
+  init_args:
+    augs_neg:
+      enable: false
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth: 0
+      pitch_range_parselmouth: 0
+      pitch_shift_parselmouth_prob: 0
+    positive_examples: same_clip
+    dataset_dirs:
+    - tencys_vocals
+    - ghero_vocals_3
+    - ghero_vocals_4
+    batch_size: 55
+    batch_size_val: 55
+    nr_samples: 176000
+    normalize: true
+    num_workers: 40
+    sr: 44100
+    batch_sampling_mode: sample_clips
+    eval_frac: 0.11
+    group_name_is_folder: true
+    group_by_artist: true
+    augs:
+      enable: true
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0.5
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth:
+      - 1
+      - 1.3
+      pitch_range_parselmouth: 1.5
+      pitch_shift_parselmouth_prob: 0.5
+    transform_override: false
+    verbose: true
+    use_random_loader: false
+    max_groups: -1
+    multi_epoch: 1
+    classification: false

unif-align/encoder.yaml

@@ -0,0 +1,3 @@
+backbone: efficientnet_b0
+embedding_dim: 1000
+pretrained: true

unif-align/feature_extractor.yaml

@@ -0,0 +1,3 @@
+hop_length: 512
+n_fft: 2048
+spec_layer: melspectogram

unif-align/hyperparams.yaml

@@ -0,0 +1,8 @@
+encoder:
+  backbone: efficientnet_b0
+  embedding_dim: 1000
+  pretrained: true
+feature_extractor:
+  hop_length: 512
+  n_fft: 2048
+  spec_layer: melspectogram

unif-align/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd51dde0840cf75ac049c11f631f738a636395f0493f6e8508f9bb36ca73db31
+size 38765709

unif-align/model.ts

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8b2ebb000ed138b0d92fa25248c7dc91e9a7189ce3e0600121791dfca496cab
+size 38970610

vicreg/config.yaml

@@ -0,0 +1,265 @@
+# pytorch_lightning==1.9.3
+seed_everything: 123
+trainer:
+  logger:
+    class_path: pytorch_lightning.loggers.TensorBoardLogger
+    init_args:
+      save_dir: logs
+      name: exp_vicreg_sameclip
+      version: null
+      log_graph: false
+      default_hp_metric: true
+      prefix: ''
+      sub_dir: null
+  enable_checkpointing: true
+  callbacks:
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-val-loss-{epoch}-{step}
+      monitor: loss/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-eer-val-{epoch}-{step}
+      monitor: EER evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-rank-val-{epoch}-{step}
+      monitor: Order evaluation mean proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-alignment-val-{epoch}-{step}
+      monitor: Alignment evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-uniformity-val-{epoch}-{step}
+      monitor: Uniformity evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: cptk-{epoch}-{step}
+      monitor: null
+      verbose: false
+      save_last: null
+      save_top_k: -1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: 25
+      save_on_train_epoch_end: null
+  - class_path: callbacks.evaluation.OrderEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  - class_path: callbacks.evaluation.EEREvaluation
+    init_args:
+      use_more_neg: false
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  - class_path: callbacks.evaluation.HypersphereEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  default_root_dir: null
+  gradient_clip_val: null
+  gradient_clip_algorithm: null
+  num_nodes: 1
+  num_processes: null
+  devices: null
+  gpus: 2
+  auto_select_gpus: null
+  tpu_cores: null
+  ipus: null
+  enable_progress_bar: true
+  overfit_batches: 0.0
+  track_grad_norm: -1
+  check_val_every_n_epoch: 1
+  fast_dev_run: false
+  accumulate_grad_batches: null
+  max_epochs: 100000
+  min_epochs: null
+  max_steps: 1000000000
+  min_steps: null
+  max_time: null
+  limit_train_batches: null
+  limit_val_batches: null
+  limit_test_batches: null
+  limit_predict_batches: null
+  val_check_interval: null
+  log_every_n_steps: 50
+  accelerator: gpu
+  strategy: null
+  sync_batchnorm: false
+  precision: 32
+  enable_model_summary: true
+  num_sanity_val_steps: 2
+  resume_from_checkpoint: null
+  profiler: null
+  benchmark: null
+  deterministic: null
+  reload_dataloaders_every_n_epochs: 0
+  auto_lr_find: false
+  replace_sampler_ddp: true
+  detect_anomaly: false
+  auto_scale_batch_size: false
+  plugins: null
+  amp_backend: null
+  amp_level: null
+  move_metrics_to_cpu: false
+  multiple_trainloader_mode: max_size_cycle
+  inference_mode: true
+ckpt_path: null
+model:
+  class_path: models.trainer.ContrastiveTrainer
+  init_args:
+    feature_extractor:
+      spec_layer: melspectogram
+      n_fft: 2048
+      hop_length: 512
+    backbone:
+      backbone: efficientnet_b0
+      pretrained: true
+      embedding_dim: 1000
+    projection:
+      input_dim: 1000
+      output_dim: 128
+      l2_normalize: true
+    optimizer1_init:
+      class_path: torch.optim.Adam
+      init_args:
+        lr: 0.0001
+        weight_decay: 1.0e-05
+    use_contrastive_loss: false
+    temp: 0.1
+    nr_negative: 64
+    decouple: false
+    use_norm_reg: false
+    max_norm_hinge: 4.0
+    norm_hinge_fact: 10.0
+    use_invariance_loss: true
+    fact_inv_loss: 25.0
+    use_covariance_reg: true
+    fact_cov: 100.0
+    use_variance_reg: true
+    fact_var: 25.0
+    gamma: 1.0
+    use_vicreg_loss: false
+    use_align_loss: false
+    fact_align_loss: 0.25
+    fact_unif_loss: 0.5
+    use_uniform_loss: false
+    mask_batch: false
+    compute_test_loss: false
+data:
+  class_path: data.vocals.VocalsDataModule
+  init_args:
+    augs_neg:
+      enable: false
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth: 0
+      pitch_range_parselmouth: 0
+      pitch_shift_parselmouth_prob: 0
+    positive_examples: same_clip
+    dataset_dirs:
+    - tencys_vocals
+    - ghero_vocals_3
+    - ghero_vocals_4
+    batch_size: 55
+    batch_size_val: 55
+    nr_samples: 176000
+    normalize: true
+    num_workers: 40
+    sr: 44100
+    batch_sampling_mode: sample_clips
+    eval_frac: 0.11
+    group_name_is_folder: true
+    group_by_artist: true
+    augs:
+      enable: true
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0.5
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth:
+      - 1
+      - 1.3
+      pitch_range_parselmouth: 1.5
+      pitch_shift_parselmouth_prob: 0.5
+    transform_override: false
+    verbose: true
+    use_random_loader: false
+    max_groups: -1
+    multi_epoch: 1
+    classification: false

vicreg/encoder.yaml

@@ -0,0 +1,3 @@
+backbone: efficientnet_b0
+embedding_dim: 1000
+pretrained: true

vicreg/feature_extractor.yaml

@@ -0,0 +1,3 @@
+hop_length: 512
+n_fft: 2048
+spec_layer: melspectogram

vicreg/hyperparams.yaml

@@ -0,0 +1,8 @@
+encoder:
+  backbone: efficientnet_b0
+  embedding_dim: 1000
+  pretrained: true
+feature_extractor:
+  hop_length: 512
+  n_fft: 2048
+  spec_layer: melspectogram

vicreg/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bccb3c2ec820feb16f109d33439886b77c8b0a236267f0ff71276ad562171cae
+size 38765709

vicreg/model.ts

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70d2c1af904ee08cb2c163307fc54c5381774774a42c96eeb2896f475e2c8aeb
+size 38970610