From 914e5958d168355195f2eec0169da687395bde2a Mon Sep 17 00:00:00 2001 From: Gustaf Rydholm Date: Sun, 24 Oct 2021 00:58:27 +0200 Subject: Rename config from cnn... to conv... --- .../experiment/cnn_transformer_paragraphs.yaml | 149 -------------------- .../experiment/cnn_transformer_paragraphs_wp.yaml | 153 --------------------- .../experiment/conv_transformer_paragraphs.yaml | 149 ++++++++++++++++++++ .../experiment/conv_transformer_paragraphs_wp.yaml | 153 +++++++++++++++++++++ 4 files changed, 302 insertions(+), 302 deletions(-) delete mode 100644 training/conf/experiment/cnn_transformer_paragraphs.yaml delete mode 100644 training/conf/experiment/cnn_transformer_paragraphs_wp.yaml create mode 100644 training/conf/experiment/conv_transformer_paragraphs.yaml create mode 100644 training/conf/experiment/conv_transformer_paragraphs_wp.yaml (limited to 'training/conf') diff --git a/training/conf/experiment/cnn_transformer_paragraphs.yaml b/training/conf/experiment/cnn_transformer_paragraphs.yaml deleted file mode 100644 index 5ee5536..0000000 --- a/training/conf/experiment/cnn_transformer_paragraphs.yaml +++ /dev/null @@ -1,149 +0,0 @@ -# @package _global_ - -defaults: - - override /mapping: null - - override /criterion: null - - override /callbacks: htr - - override /datamodule: iam_extended_paragraphs - - override /network: null - - override /model: null - - override /lr_schedulers: null - - override /optimizers: null - - -epochs: &epochs 1000 -ignore_index: &ignore_index 3 -num_classes: &num_classes 58 -max_output_len: &max_output_len 682 -summary: [[1, 1, 576, 640], [1, 682]] - -criterion: - _target_: torch.nn.CrossEntropyLoss - ignore_index: *ignore_index - -mapping: &mapping - mapping: - _target_: text_recognizer.data.mappings.emnist_mapping.EmnistMapping - extra_symbols: [ "\n" ] - -callbacks: - stochastic_weight_averaging: - _target_: pytorch_lightning.callbacks.StochasticWeightAveraging - swa_epoch_start: 0.75 - swa_lrs: 1.0e-5 - annealing_epochs: 10 - annealing_strategy: cos - device: null - -optimizers: - madgrad: - _target_: madgrad.MADGRAD - lr: 3.0e-4 - momentum: 0.9 - weight_decay: 5.0e-6 - eps: 1.0e-6 - - parameters: network - -lr_schedulers: - network: - _target_: torch.optim.lr_scheduler.OneCycleLR - max_lr: 3.0e-4 - total_steps: null - epochs: *epochs - steps_per_epoch: 632 - pct_start: 0.03 - anneal_strategy: cos - cycle_momentum: true - base_momentum: 0.85 - max_momentum: 0.95 - div_factor: 25 - final_div_factor: 1.0e4 - three_phase: false - last_epoch: -1 - verbose: false - # Non-class arguments - interval: step - monitor: val/loss - -datamodule: - _target_: text_recognizer.data.iam_extended_paragraphs.IAMExtendedParagraphs - batch_size: 4 - num_workers: 12 - train_fraction: 0.8 - pin_memory: true - << : *mapping - -network: - _target_: text_recognizer.networks.conv_transformer.ConvTransformer - input_dims: [1, 576, 640] - hidden_dim: &hidden_dim 128 - encoder_dim: 1280 - dropout_rate: 0.2 - num_classes: *num_classes - pad_index: *ignore_index - encoder: - _target_: text_recognizer.networks.encoders.efficientnet.EfficientNet - arch: b0 - out_channels: 1280 - stochastic_dropout_rate: 0.2 - bn_momentum: 0.99 - bn_eps: 1.0e-3 - decoder: - _target_: text_recognizer.networks.transformer.Decoder - dim: *hidden_dim - depth: 3 - num_heads: 4 - attn_fn: text_recognizer.networks.transformer.attention.Attention - attn_kwargs: - dim_head: 32 - dropout_rate: 0.2 - norm_fn: text_recognizer.networks.transformer.norm.ScaleNorm - ff_fn: text_recognizer.networks.transformer.mlp.FeedForward - ff_kwargs: - dim_out: null - expansion_factor: 4 - glu: true - dropout_rate: 0.2 - cross_attend: true - pre_norm: true - rotary_emb: - _target_: text_recognizer.networks.transformer.positional_encodings.rotary_embedding.RotaryEmbedding - dim: 32 - pixel_pos_embedding: - _target_: text_recognizer.networks.transformer.positional_encodings.PositionalEncoding2D - hidden_dim: *hidden_dim - max_h: 18 - max_w: 20 - token_pos_embedding: - _target_: text_recognizer.networks.transformer.positional_encodings.PositionalEncoding - hidden_dim: *hidden_dim - dropout_rate: 0.2 - max_len: *max_output_len - -model: - _target_: text_recognizer.models.transformer.TransformerLitModel - << : *mapping - max_output_len: *max_output_len - start_token: - end_token: - pad_token:

- -trainer: - _target_: pytorch_lightning.Trainer - stochastic_weight_avg: true - auto_scale_batch_size: binsearch - auto_lr_find: false - gradient_clip_val: 0.0 - fast_dev_run: false - gpus: 1 - precision: 16 - max_epochs: *epochs - terminate_on_nan: true - weights_summary: null - limit_train_batches: 1.0 - limit_val_batches: 1.0 - limit_test_batches: 1.0 - resume_from_checkpoint: null - accumulate_grad_batches: 16 - overfit_batches: 0 diff --git a/training/conf/experiment/cnn_transformer_paragraphs_wp.yaml b/training/conf/experiment/cnn_transformer_paragraphs_wp.yaml deleted file mode 100644 index 499a609..0000000 --- a/training/conf/experiment/cnn_transformer_paragraphs_wp.yaml +++ /dev/null @@ -1,153 +0,0 @@ -defaults: - - override /mapping: null - - override /criterion: null - - override /datamodule: null - - override /network: null - - override /model: null - - override /lr_schedulers: null - - override /optimizers: null - - -epochs: &epochs 1000 -ignore_index: &ignore_index 1000 -num_classes: &num_classes 1006 -max_output_len: &max_output_len 451 -summary: [[1, 1, 576, 640], [1, 451]] - -criterion: - _target_: torch.nn.CrossEntropyLoss - ignore_index: *ignore_index - -mapping: - _target_: text_recognizer.data.word_piece_mapping.WordPieceMapping - num_features: 1000 - tokens: iamdb_1kwp_tokens_1000.txt - lexicon: iamdb_1kwp_lex_1000.txt - data_dir: null - use_words: false - prepend_wordsep: false - special_tokens: [ , ,

] - extra_symbols: [ "\n" ] - -callbacks: - stochastic_weight_averaging: - _target_: pytorch_lightning.callbacks.StochasticWeightAveraging - swa_epoch_start: 0.75 - swa_lrs: 1.0e-5 - annealing_epochs: 10 - annealing_strategy: cos - device: null - -optimizers: - madgrad: - _target_: madgrad.MADGRAD - lr: 3.0e-4 - momentum: 0.9 - weight_decay: 0 - eps: 1.0e-6 - - parameters: network - -lr_schedulers: - network: - _target_: torch.optim.lr_scheduler.OneCycleLR - max_lr: 3.0e-4 - total_steps: null - epochs: *epochs - steps_per_epoch: 632 - pct_start: 0.3 - anneal_strategy: cos - cycle_momentum: true - base_momentum: 0.85 - max_momentum: 0.95 - div_factor: 25 - final_div_factor: 1.0e4 - three_phase: false - last_epoch: -1 - verbose: false - # Non-class arguments - interval: step - monitor: val/loss - -datamodule: - _target_: text_recognizer.data.iam_extended_paragraphs.IAMExtendedParagraphs - batch_size: 4 - num_workers: 12 - train_fraction: 0.8 - augment: true - pin_memory: true - word_pieces: true - resize: null - -network: - _target_: text_recognizer.networks.conv_transformer.ConvTransformer - input_dims: [1, 576, 640] - hidden_dim: &hidden_dim 128 - encoder_dim: 1280 - dropout_rate: 0.2 - num_classes: *num_classes - pad_index: *ignore_index - encoder: - _target_: text_recognizer.networks.encoders.efficientnet.EfficientNet - arch: b0 - out_channels: 1280 - stochastic_dropout_rate: 0.2 - bn_momentum: 0.99 - bn_eps: 1.0e-3 - decoder: - _target_: text_recognizer.networks.transformer.Decoder - dim: *hidden_dim - depth: 3 - num_heads: 4 - attn_fn: text_recognizer.networks.transformer.attention.Attention - attn_kwargs: - dim_head: 32 - dropout_rate: 0.2 - norm_fn: text_recognizer.networks.transformer.norm.ScaleNorm - ff_fn: text_recognizer.networks.transformer.mlp.FeedForward - ff_kwargs: - dim_out: null - expansion_factor: 4 - glu: true - dropout_rate: 0.2 - cross_attend: true - pre_norm: true - rotary_emb: - _target_: text_recognizer.networks.transformer.positional_encodings.rotary_embedding.RotaryEmbedding - dim: 32 - pixel_pos_embedding: - _target_: text_recognizer.networks.transformer.positional_encodings.PositionalEncoding2D - hidden_dim: *hidden_dim - max_h: 18 - max_w: 20 - token_pos_embedding: - _target_: text_recognizer.networks.transformer.positional_encodings.PositionalEncoding - hidden_dim: *hidden_dim - dropout_rate: 0.2 - max_len: *max_output_len - -model: - _target_: text_recognizer.models.transformer.TransformerLitModel - max_output_len: *max_output_len - start_token: - end_token: - pad_token:

- -trainer: - _target_: pytorch_lightning.Trainer - stochastic_weight_avg: true - auto_scale_batch_size: binsearch - auto_lr_find: false - gradient_clip_val: 0.0 - fast_dev_run: false - gpus: 1 - precision: 16 - max_epochs: *epochs - terminate_on_nan: true - weights_summary: null - limit_train_batches: 1.0 - limit_val_batches: 1.0 - limit_test_batches: 1.0 - resume_from_checkpoint: null - accumulate_grad_batches: 16 - overfit_batches: 0 diff --git a/training/conf/experiment/conv_transformer_paragraphs.yaml b/training/conf/experiment/conv_transformer_paragraphs.yaml new file mode 100644 index 0000000..ebbd6ef --- /dev/null +++ b/training/conf/experiment/conv_transformer_paragraphs.yaml @@ -0,0 +1,149 @@ +# @package _global_ + +defaults: + - override /mapping: null + - override /criterion: null + - override /callbacks: htr + - override /datamodule: iam_extended_paragraphs + - override /network: null + - override /model: null + - override /lr_schedulers: null + - override /optimizers: null + + +epochs: &epochs 1000 +ignore_index: &ignore_index 3 +num_classes: &num_classes 58 +max_output_len: &max_output_len 682 +summary: [[1, 1, 576, 640], [1, 682]] + +criterion: + _target_: torch.nn.CrossEntropyLoss + ignore_index: *ignore_index + +mapping: &mapping + mapping: + _target_: text_recognizer.data.mappings.emnist_mapping.EmnistMapping + extra_symbols: [ "\n" ] + +callbacks: + stochastic_weight_averaging: + _target_: pytorch_lightning.callbacks.StochasticWeightAveraging + swa_epoch_start: 0.75 + swa_lrs: 1.0e-5 + annealing_epochs: 10 + annealing_strategy: cos + device: null + +optimizers: + madgrad: + _target_: madgrad.MADGRAD + lr: 2.0e-4 + momentum: 0.9 + weight_decay: 5.0e-6 + eps: 1.0e-6 + + parameters: network + +lr_schedulers: + network: + _target_: torch.optim.lr_scheduler.OneCycleLR + max_lr: 2.0e-4 + total_steps: null + epochs: *epochs + steps_per_epoch: 632 + pct_start: 0.03 + anneal_strategy: cos + cycle_momentum: true + base_momentum: 0.85 + max_momentum: 0.95 + div_factor: 25 + final_div_factor: 1.0e4 + three_phase: false + last_epoch: -1 + verbose: false + # Non-class arguments + interval: step + monitor: val/loss + +datamodule: + _target_: text_recognizer.data.iam_extended_paragraphs.IAMExtendedParagraphs + batch_size: 4 + num_workers: 12 + train_fraction: 0.8 + pin_memory: true + << : *mapping + +network: + _target_: text_recognizer.networks.conv_transformer.ConvTransformer + input_dims: [1, 576, 640] + hidden_dim: &hidden_dim 128 + encoder_dim: 1280 + dropout_rate: 0.2 + num_classes: *num_classes + pad_index: *ignore_index + encoder: + _target_: text_recognizer.networks.encoders.efficientnet.EfficientNet + arch: b0 + out_channels: 1280 + stochastic_dropout_rate: 0.2 + bn_momentum: 0.99 + bn_eps: 1.0e-3 + decoder: + _target_: text_recognizer.networks.transformer.Decoder + dim: *hidden_dim + depth: 3 + num_heads: 4 + attn_fn: text_recognizer.networks.transformer.attention.Attention + attn_kwargs: + dim_head: 32 + dropout_rate: 0.2 + norm_fn: text_recognizer.networks.transformer.norm.ScaleNorm + ff_fn: text_recognizer.networks.transformer.mlp.FeedForward + ff_kwargs: + dim_out: null + expansion_factor: 4 + glu: true + dropout_rate: 0.2 + cross_attend: true + pre_norm: true + rotary_emb: + _target_: text_recognizer.networks.transformer.positional_encodings.rotary_embedding.RotaryEmbedding + dim: 32 + pixel_pos_embedding: + _target_: text_recognizer.networks.transformer.positional_encodings.PositionalEncoding2D + hidden_dim: *hidden_dim + max_h: 18 + max_w: 20 + token_pos_embedding: + _target_: text_recognizer.networks.transformer.positional_encodings.PositionalEncoding + hidden_dim: *hidden_dim + dropout_rate: 0.2 + max_len: *max_output_len + +model: + _target_: text_recognizer.models.transformer.TransformerLitModel + << : *mapping + max_output_len: *max_output_len + start_token: + end_token: + pad_token:

+ +trainer: + _target_: pytorch_lightning.Trainer + stochastic_weight_avg: true + auto_scale_batch_size: binsearch + auto_lr_find: false + gradient_clip_val: 0.75 + fast_dev_run: false + gpus: 1 + precision: 16 + max_epochs: *epochs + terminate_on_nan: true + weights_summary: null + limit_train_batches: 1.0 + limit_val_batches: 1.0 + limit_test_batches: 1.0 + resume_from_checkpoint: null + accumulate_grad_batches: 16 + overfit_batches: 0 diff --git a/training/conf/experiment/conv_transformer_paragraphs_wp.yaml b/training/conf/experiment/conv_transformer_paragraphs_wp.yaml new file mode 100644 index 0000000..499a609 --- /dev/null +++ b/training/conf/experiment/conv_transformer_paragraphs_wp.yaml @@ -0,0 +1,153 @@ +defaults: + - override /mapping: null + - override /criterion: null + - override /datamodule: null + - override /network: null + - override /model: null + - override /lr_schedulers: null + - override /optimizers: null + + +epochs: &epochs 1000 +ignore_index: &ignore_index 1000 +num_classes: &num_classes 1006 +max_output_len: &max_output_len 451 +summary: [[1, 1, 576, 640], [1, 451]] + +criterion: + _target_: torch.nn.CrossEntropyLoss + ignore_index: *ignore_index + +mapping: + _target_: text_recognizer.data.word_piece_mapping.WordPieceMapping + num_features: 1000 + tokens: iamdb_1kwp_tokens_1000.txt + lexicon: iamdb_1kwp_lex_1000.txt + data_dir: null + use_words: false + prepend_wordsep: false + special_tokens: [ , ,

] + extra_symbols: [ "\n" ] + +callbacks: + stochastic_weight_averaging: + _target_: pytorch_lightning.callbacks.StochasticWeightAveraging + swa_epoch_start: 0.75 + swa_lrs: 1.0e-5 + annealing_epochs: 10 + annealing_strategy: cos + device: null + +optimizers: + madgrad: + _target_: madgrad.MADGRAD + lr: 3.0e-4 + momentum: 0.9 + weight_decay: 0 + eps: 1.0e-6 + + parameters: network + +lr_schedulers: + network: + _target_: torch.optim.lr_scheduler.OneCycleLR + max_lr: 3.0e-4 + total_steps: null + epochs: *epochs + steps_per_epoch: 632 + pct_start: 0.3 + anneal_strategy: cos + cycle_momentum: true + base_momentum: 0.85 + max_momentum: 0.95 + div_factor: 25 + final_div_factor: 1.0e4 + three_phase: false + last_epoch: -1 + verbose: false + # Non-class arguments + interval: step + monitor: val/loss + +datamodule: + _target_: text_recognizer.data.iam_extended_paragraphs.IAMExtendedParagraphs + batch_size: 4 + num_workers: 12 + train_fraction: 0.8 + augment: true + pin_memory: true + word_pieces: true + resize: null + +network: + _target_: text_recognizer.networks.conv_transformer.ConvTransformer + input_dims: [1, 576, 640] + hidden_dim: &hidden_dim 128 + encoder_dim: 1280 + dropout_rate: 0.2 + num_classes: *num_classes + pad_index: *ignore_index + encoder: + _target_: text_recognizer.networks.encoders.efficientnet.EfficientNet + arch: b0 + out_channels: 1280 + stochastic_dropout_rate: 0.2 + bn_momentum: 0.99 + bn_eps: 1.0e-3 + decoder: + _target_: text_recognizer.networks.transformer.Decoder + dim: *hidden_dim + depth: 3 + num_heads: 4 + attn_fn: text_recognizer.networks.transformer.attention.Attention + attn_kwargs: + dim_head: 32 + dropout_rate: 0.2 + norm_fn: text_recognizer.networks.transformer.norm.ScaleNorm + ff_fn: text_recognizer.networks.transformer.mlp.FeedForward + ff_kwargs: + dim_out: null + expansion_factor: 4 + glu: true + dropout_rate: 0.2 + cross_attend: true + pre_norm: true + rotary_emb: + _target_: text_recognizer.networks.transformer.positional_encodings.rotary_embedding.RotaryEmbedding + dim: 32 + pixel_pos_embedding: + _target_: text_recognizer.networks.transformer.positional_encodings.PositionalEncoding2D + hidden_dim: *hidden_dim + max_h: 18 + max_w: 20 + token_pos_embedding: + _target_: text_recognizer.networks.transformer.positional_encodings.PositionalEncoding + hidden_dim: *hidden_dim + dropout_rate: 0.2 + max_len: *max_output_len + +model: + _target_: text_recognizer.models.transformer.TransformerLitModel + max_output_len: *max_output_len + start_token: + end_token: + pad_token:

+ +trainer: + _target_: pytorch_lightning.Trainer + stochastic_weight_avg: true + auto_scale_batch_size: binsearch + auto_lr_find: false + gradient_clip_val: 0.0 + fast_dev_run: false + gpus: 1 + precision: 16 + max_epochs: *epochs + terminate_on_nan: true + weights_summary: null + limit_train_batches: 1.0 + limit_val_batches: 1.0 + limit_test_batches: 1.0 + resume_from_checkpoint: null + accumulate_grad_batches: 16 + overfit_batches: 0 -- cgit v1.2.3-70-g09d2