summaryrefslogtreecommitdiff
path: root/training/conf
diff options
context:
space:
mode:
authorGustaf Rydholm <gustaf.rydholm@gmail.com>2021-11-22 22:38:43 +0100
committerGustaf Rydholm <gustaf.rydholm@gmail.com>2021-11-22 22:38:43 +0100
commit050e1bd284a173d2586ad4607e95d114691db563 (patch)
treef428b5a17396a7cd585e89d84765e3d7ed233618 /training/conf
parent36875ceca1e00f5bb39a151c50ecf5e333b4cf79 (diff)
Move efficientnet from encoder dir
Diffstat (limited to 'training/conf')
-rw-r--r--training/conf/experiment/conv_transformer_lines.yaml11
-rw-r--r--training/conf/experiment/conv_transformer_paragraphs.yaml2
-rw-r--r--training/conf/experiment/conv_transformer_paragraphs_wp.yaml2
-rw-r--r--training/conf/network/encoder/efficientnet.yaml2
4 files changed, 9 insertions, 8 deletions
diff --git a/training/conf/experiment/conv_transformer_lines.yaml b/training/conf/experiment/conv_transformer_lines.yaml
index 6c266b8..2918317 100644
--- a/training/conf/experiment/conv_transformer_lines.yaml
+++ b/training/conf/experiment/conv_transformer_lines.yaml
@@ -64,10 +64,10 @@ rotary_embedding: &rotary_embedding
dim: 64
attn: &attn
- dim: &hidden_dim 256
+ dim: &hidden_dim 128
num_heads: 4
dim_head: 64
- dropout_rate: &dropout_rate 0.5
+ dropout_rate: &dropout_rate 0.2
network:
_target_: text_recognizer.networks.conv_transformer.ConvTransformer
@@ -76,8 +76,9 @@ network:
num_classes: *num_classes
pad_index: *ignore_index
encoder:
- _target_: text_recognizer.networks.encoders.efficientnet.EfficientNet
- arch: b1
+ _target_: text_recognizer.networks.efficientnet.EfficientNet
+ arch: b0
+ depth: 5
stochastic_dropout_rate: 0.2
bn_momentum: 0.99
bn_eps: 1.0e-3
@@ -113,7 +114,7 @@ network:
dim: *hidden_dim
heads: 4
shape: *shape
- depth: 1
+ depth: 2
dim_head: 64
dim_index: 1
diff --git a/training/conf/experiment/conv_transformer_paragraphs.yaml b/training/conf/experiment/conv_transformer_paragraphs.yaml
index 4f15ef2..32f5763 100644
--- a/training/conf/experiment/conv_transformer_paragraphs.yaml
+++ b/training/conf/experiment/conv_transformer_paragraphs.yaml
@@ -78,7 +78,7 @@ network:
num_classes: *num_classes
pad_index: *ignore_index
encoder:
- _target_: text_recognizer.networks.encoders.efficientnet.EfficientNet
+ _target_: text_recognizer.networks.efficientnet.EfficientNet
arch: b1
stochastic_dropout_rate: 0.2
bn_momentum: 0.99
diff --git a/training/conf/experiment/conv_transformer_paragraphs_wp.yaml b/training/conf/experiment/conv_transformer_paragraphs_wp.yaml
index 91fba9a..bf192ec 100644
--- a/training/conf/experiment/conv_transformer_paragraphs_wp.yaml
+++ b/training/conf/experiment/conv_transformer_paragraphs_wp.yaml
@@ -89,7 +89,7 @@ network:
num_classes: *num_classes
pad_index: *ignore_index
encoder:
- _target_: text_recognizer.networks.encoders.efficientnet.EfficientNet
+ _target_: text_recognizer.networks.efficientnet.EfficientNet
arch: b0
out_channels: 1280
stochastic_dropout_rate: 0.2
diff --git a/training/conf/network/encoder/efficientnet.yaml b/training/conf/network/encoder/efficientnet.yaml
index 0e62293..a7be069 100644
--- a/training/conf/network/encoder/efficientnet.yaml
+++ b/training/conf/network/encoder/efficientnet.yaml
@@ -1,4 +1,4 @@
-_target_: text_recognizer.networks.encoders.efficientnet.EfficientNet
+_target_: text_recognizer.networks.efficientnet.EfficientNet
arch: b0
stochastic_dropout_rate: 0.2
bn_momentum: 0.99