tf.nn.conv2d vs tf.layers.conv2d

The main difference is that filter(s).

• tf.layers.conv2d : It handles activation and bias automatically
• tf.nn.conv2d : This takes tensor as a filter, Hence you can specify the weight decay. like the following cifar10 code

kernel = _variable_with_weight_decay('weights',
                                     shape=[5, 5, 3, 64],
                                     stddev=5e-2,
                                     wd=0.0)
conv = tf.nn.conv2d(images, kernel, [1, 1, 1, 1], padding='SAME')