diff options
Diffstat (limited to 'training/conf/network')
| -rw-r--r-- | training/conf/network/decoder/pixelcnn_encoder.yaml | 5 | ||||
| -rw-r--r-- | training/conf/network/decoder/vae_decoder.yaml | 5 | ||||
| -rw-r--r-- | training/conf/network/encoder/pixelcnn_decoder.yaml | 5 | ||||
| -rw-r--r-- | training/conf/network/encoder/vae_encoder.yaml | 5 | ||||
| -rw-r--r-- | training/conf/network/vqvae.yaml | 15 | ||||
| -rw-r--r-- | training/conf/network/vqvae_pixelcnn.yaml | 9 |
6 files changed, 37 insertions, 7 deletions
diff --git a/training/conf/network/decoder/pixelcnn_encoder.yaml b/training/conf/network/decoder/pixelcnn_encoder.yaml new file mode 100644 index 0000000..47a130d --- /dev/null +++ b/training/conf/network/decoder/pixelcnn_encoder.yaml @@ -0,0 +1,5 @@ +_target_: text_recognizer.networks.vqvae.pixelcnn.Encoder +in_channels: 1 +hidden_dim: 8 +channels_multipliers: [1, 2, 8, 8] +dropout_rate: 0.25 diff --git a/training/conf/network/decoder/vae_decoder.yaml b/training/conf/network/decoder/vae_decoder.yaml new file mode 100644 index 0000000..b2090b3 --- /dev/null +++ b/training/conf/network/decoder/vae_decoder.yaml @@ -0,0 +1,5 @@ +_target_: text_recognizer.networks.vqvae.decoder.Decoder +out_channels: 1 +hidden_dim: 32 +channels_multipliers: [8, 6, 2, 1] +dropout_rate: 0.25 diff --git a/training/conf/network/encoder/pixelcnn_decoder.yaml b/training/conf/network/encoder/pixelcnn_decoder.yaml new file mode 100644 index 0000000..3895164 --- /dev/null +++ b/training/conf/network/encoder/pixelcnn_decoder.yaml @@ -0,0 +1,5 @@ +_target_: text_recognizer.networks.vqvae.pixelcnn.Decoder +out_channels: 1 +hidden_dim: 8 +channels_multipliers: [8, 8, 2, 1] +dropout_rate: 0.25 diff --git a/training/conf/network/encoder/vae_encoder.yaml b/training/conf/network/encoder/vae_encoder.yaml new file mode 100644 index 0000000..5dc6814 --- /dev/null +++ b/training/conf/network/encoder/vae_encoder.yaml @@ -0,0 +1,5 @@ +_target_: text_recognizer.networks.vqvae.encoder.Encoder +in_channels: 1 +hidden_dim: 32 +channels_multipliers: [1, 2, 6, 8] +dropout_rate: 0.25 diff --git a/training/conf/network/vqvae.yaml b/training/conf/network/vqvae.yaml index 5a5c066..835d0b7 100644 --- a/training/conf/network/vqvae.yaml +++ b/training/conf/network/vqvae.yaml @@ -1,8 +1,9 @@ -_target_: text_recognizer.networks.vqvae.VQVAE -in_channels: 1 -res_channels: 32 -num_residual_layers: 2 -embedding_dim: 64 -num_embeddings: 512 +defaults: + - encoder: vae_encoder + - decoder: vae_decoder + +_target_: text_recognizer.networks.vqvae.vqvae.VQVAE +hidden_dim: 256 +embedding_dim: 32 +num_embeddings: 1024 decay: 0.99 -activation: mish diff --git a/training/conf/network/vqvae_pixelcnn.yaml b/training/conf/network/vqvae_pixelcnn.yaml new file mode 100644 index 0000000..10200bc --- /dev/null +++ b/training/conf/network/vqvae_pixelcnn.yaml @@ -0,0 +1,9 @@ +defaults: + - encoder: pixelcnn_encoder + - decoder: pixelcnn_decoder + +_target_: text_recognizer.networks.vqvae.vqvae.VQVAE +hidden_dim: 64 +embedding_dim: 32 +num_embeddings: 512 +decay: 0.99 |