把卷积隔离开，只训练卷积层

load.py

@@ -55,4 +55,13 @@ def get_batch_data(train_data, train_label,index,batch_size):
 filename = 'dataset.npy'
 dataset = read_dataset(filename)
 train_data, train_label = get_train_data(dataset)
-print('==========================加载数据成功===========================')
+dataset1 = []
+for i in range(3200):
+    data = train_data[i]
+    label = train_label[i]
+    data = data / 127.5 - 1
+    dataset1.append([data,label])
+#print(dataset1)
+np.save("train.npy",dataset1)
+#print(len(train_data))
+#print('==========================加载数据成功===========================')

net.py

@@ -8,7 +8,7 @@ def __init__(self, layer_names, weights):
         self.layer_names = layer_names
         self.weights = weights
         self.alpha = 0.1
-        self.batch_size = 32
+        self.batch_size = 8
         self.step = 100
         self.build()
         self.set_training()
@@ -121,11 +121,11 @@ def loss(self, result, label, size):
         prob_pre =  tf.slice(self.result, [0, 7*7], [size, 7*7], name='prob_pre')
         prob_label =  tf.slice(label, [0, 7*7], [size, 7*7], name='prob_label')
         # 预测横纵坐标
-        pos_pre = tf.slice(self.result, [0, 7*7*2], [size, 7*7*4], name='pos_pre')
-        pos_label = tf.slice(label, [0, 7*7*2], [size, 7*7*4], name='pos_label')
+        pos_pre = tf.slice(self.result, [0, 7*7*2], [size, 7*7*2], name='pos_pre')
+        pos_label = tf.slice(label, [0, 7*7*2], [size, 7*7*2], name='pos_label')
         # 预测长宽尺寸
-        size_pre = tf.slice(self.result, [0, 7*7*4], [size, 7*7*6], name='size_pre')
-        size_label = tf.slice(label, [0, 7*7*4], [size, 7*7*6], name='size_label')
+        size_pre = tf.slice(self.result, [0, 7*7*4], [size, 7*7*2], name='size_pre')
+        size_label = tf.slice(label, [0, 7*7*4], [size, 7*7*2], name='size_label')
         # 损失
         class_loss = tf.reduce_sum(tf.square( class_pre - class_label ))
         prob_loss = tf.reduce_sum(tf.square( prob_pre - prob_label ))