rename head -> gears

ppcls/arch/__init__.py

@@ -18,10 +18,10 @@ import importlib
 import paddle.nn as nn
 
 from . import backbone
-from . import head
+from . import gears
 
 from .backbone import *
-from .head import *
+from .gears import *
 from .utils import *
 
 __all__ = ["build_model", "RecModel"]

ppcls/arch/head/__init__.py → ppcls/arch/gears/__init__.py

@@ -19,10 +19,11 @@ from .fc import FC
 
 __all__ = ['build_head']
 
+
 def build_head(config):
     support_dict = ['ArcMargin', 'CosMargin', 'CircleMargin', 'FC']
     module_name = config.pop('name')
-    assert module_name in support_dict, Exception('head only support {}'.format(
-        support_dict))
+    assert module_name in support_dict, Exception(
+        'head only support {}'.format(support_dict))
     module_class = eval(module_name)(**config)
     return module_class

ppcls/arch/head/circlemargin.py → ppcls/arch/gears/circlemargin.py

@@ -16,30 +16,32 @@ import math
 import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
- 
+
+
 class CircleMargin(nn.Layer):
-    def __init__(self, embedding_size, 
-                       class_num, 
-                       margin, 
-                       scale):
+    def __init__(self, embedding_size, class_num, margin, scale):
         super(CircleSoftmax, self).__init__()
-        self.scale  = scale
+        self.scale = scale
         self.margin = margin
         self.embedding_size = embedding_size
         self.class_num = class_num
 
-        weight_attr = paddle.ParamAttr(initializer = paddle.nn.initializer.XavierNormal())
-        self.fc0 = paddle.nn.Linear(self.embedding_size, self.class_num, weight_attr=weight_attr)
- 
+        weight_attr = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.XavierNormal())
+        self.fc0 = paddle.nn.Linear(
+            self.embedding_size, self.class_num, weight_attr=weight_attr)
+
     def forward(self, input, label):
-        feat_norm = paddle.sqrt(paddle.sum(paddle.square(input), axis=1, keepdim=True))
+        feat_norm = paddle.sqrt(
+            paddle.sum(paddle.square(input), axis=1, keepdim=True))
         input = paddle.divide(input, feat_norm)
 
         weight = self.fc0.weight
-        weight_norm = paddle.sqrt(paddle.sum(paddle.square(weight), axis=0, keepdim=True))
+        weight_norm = paddle.sqrt(
+            paddle.sum(paddle.square(weight), axis=0, keepdim=True))
         weight = paddle.divide(weight, weight_norm)
- 
-        logits   = paddle.matmul(input, weight)
+
+        logits = paddle.matmul(input, weight)
 
         alpha_p = paddle.clip(-logits.detach() + 1 + self.margin, min=0.)
         alpha_n = paddle.clip(logits.detach() + self.margin, min=0.)
@@ -51,5 +53,5 @@ class CircleMargin(nn.Layer):
         logits_n = alpha_n * (logits - delta_n)
         pre_logits = logits_p * m_hot + logits_n * (1 - m_hot)
         pre_logits = self.scale * pre_logits
- 
+
         return pre_logits

ppcls/arch/head/cosmargin.py → ppcls/arch/gears/cosmargin.py

@@ -16,35 +16,41 @@ import paddle
 import math
 import paddle.nn as nn
 
+
 class CosMargin(paddle.nn.Layer):
-    def __init__(self, embedding_size,
-                       class_num,
-                       margin=0.35,
-                       scale=64.0):
+    def __init__(self, embedding_size, class_num, margin=0.35, scale=64.0):
         super(CosMargin, self).__init__()
         self.scale = scale
         self.margin = margin
         self.embedding_size = embedding_size
         self.class_num = class_num
-        
-        weight_attr =  paddle.ParamAttr(initializer = paddle.nn.initializer.XavierNormal())
-        self.fc = nn.Linear(self.embedding_size, self.class_num, weight_attr=weight_attr, bias_attr=False)
-        
+
+        weight_attr = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.XavierNormal())
+        self.fc = nn.Linear(
+            self.embedding_size,
+            self.class_num,
+            weight_attr=weight_attr,
+            bias_attr=False)
+
     def forward(self, input, label):
         label.stop_gradient = True
 
-        input_norm = paddle.sqrt(paddle.sum(paddle.square(input), axis=1, keepdim=True))
-        input = paddle.divide(input, x_norm) 
+        input_norm = paddle.sqrt(
+            paddle.sum(paddle.square(input), axis=1, keepdim=True))
+        input = paddle.divide(input, x_norm)
 
         weight = self.fc.weight
-        weight_norm = paddle.sqrt(paddle.sum(paddle.square(weight), axis=0, keepdim=True))
+        weight_norm = paddle.sqrt(
+            paddle.sum(paddle.square(weight), axis=0, keepdim=True))
         weight = paddle.divide(weight, weight_norm)
 
-        cos   = paddle.matmul(input, weight)
+        cos = paddle.matmul(input, weight)
         cos_m = cos - self.margin
-        
+
         one_hot = paddle.nn.functional.one_hot(label, self.class_num)
         one_hot = paddle.squeeze(one_hot, axis=[1])
-        output  = paddle.multiply(one_hot, cos_m) + paddle.multiply((1.0 - one_hot), cos)
+        output = paddle.multiply(one_hot, cos_m) + paddle.multiply(
+            (1.0 - one_hot), cos)
         output = output * self.scale
         return output

ppcls/arch/head/fc.py → ppcls/arch/gears/fc.py

@@ -19,14 +19,16 @@ from __future__ import print_function
 import paddle
 import paddle.nn as nn
 
+
 class FC(nn.Layer):
-    def __init__(self, embedding_size, 
-                       class_num):
+    def __init__(self, embedding_size, class_num):
         super(FC, self).__init__()
-        self.embedding_size  = embedding_size
+        self.embedding_size = embedding_size
         self.class_num = class_num
-        weight_attr =  paddle.ParamAttr(initializer = paddle.nn.initializer.XavierNormal())
-        self.fc  =  paddle.nn.Linear(self.embedding_size, self.class_num, weight_attr=weight_attr)    
+        weight_attr = paddle.ParamAttr(
+            initializer=paddle.nn.initializer.XavierNormal())
+        self.fc = paddle.nn.Linear(
+            self.embedding_size, self.class_num, weight_attr=weight_attr)
 
     def forward(self, input, label):
         out = self.fc(input)