documentation

keras_cv/layers/hierarchical_transformer_encoder.py

@@ -42,14 +42,14 @@ class HierarchicalTransformerEncoder(keras.layers.Layer):
             Due to the residual addition the input dimensionality has to be
             equal to the output dimensionality.
         num_heads: integer, the number of heads for the
-            `SegFormerMultiheadAttention` layer
-        drop_prob: float, default 0.0, the probability of dropping a random
-            sample using the `DropPath` layer.
-        layer_norm_epsilon: float, default 1e-06, the epsilon for
-            `LayerNormalization` layers
-        sr_ratio: integer, default 1, the ratio to use within
+            `SegFormerMultiheadAttention` layer.
+        drop_prob: float, the probability of dropping a random
+            sample using the `DropPath` layer. Defaults to `0.0`.
+        layer_norm_epsilon: float, the epsilon for
+            `LayerNormalization` layers. Defaults to `1e-06`
+        sr_ratio: integer, the ratio to use within
             `SegFormerMultiheadAttention`. If set to > 1, a `Conv2D`
-             layer is used to reduce the length of the sequence.
+             layer is used to reduce the length of the sequence. Defaults to `1`.
 
     Basic usage:
 

keras_cv/layers/overlapping_patching_embedding.py

@@ -32,10 +32,12 @@ def __init__(self, project_dim=32, patch_size=7, stride=4, **kwargs):
             - [Ported from the TensorFlow implementation from DeepVision](https://github.com/DavidLandup0/deepvision/blob/main/deepvision/layers/hierarchical_transformer_encoder.py) # noqa: E501
 
         Args:
-            project_dim: integer, default 32, the dimensionality of the projection
-            patch_size: integer, default 7, the size of the patches to encode
-            stride: integer, default 4, the stride to use for the patching before
-                projection
+            project_dim: integer, the dimensionality of the projection.
+                Defaults to `32`.
+            patch_size: integer, the size of the patches to encode.
+                Defaults to `7`.
+            stride: integer, the stride to use for the patching before
+                projection. Defaults to 5`.
 
         Basic usage:
 

keras_cv/models/backbones/mix_transformer/mix_transformer_backbone.py

@@ -69,9 +69,9 @@ def __init__(
                 values.
             num_classes: int, the number of classes for the detection model,
                 including the background class.
-            projection_filters: int, default 256, number of filters in the
+            projection_filters: int, number of filters in the
                 convolution layer projecting the concatenated features into
-                a segmentation map.
+                a segmentation map. Defaults to `256`.
 
         Examples:
 

keras_cv/models/segmentation/segformer/segformer.py

@@ -47,9 +47,9 @@ class SegFormer(Task):
             values.
         num_classes: int, the number of classes for the detection model,
             including the background class.
-        projection_filters: int, default 256, number of filters in the
+        projection_filters: int, number of filters in the
             convolution layer projecting the concatenated features into
-            a segmentation map.
+            a segmentation map. Defaults to 256`.
 
     Examples: