python - Tensorflow Depthwise Convolution Understanding -

i'm trying understand how tensorflow's depthwise convolution works. far i've understood, each channel in input image convolved it's own set of filters, , results concatenated. i'm going stick parameter depth_multiplier=1 sake of simplicity in remainder, n_inputchannels == n_outputchannels.

so in theory, split depthwise convolution n individual, regular conv2ds, correct? why following code produce different results wondering - precision issue? i'm following documentation ordering [filter_height, filter_width, in_channels, 1] depthwise convolution filters, , [filter_height, filter_width, in_channels, out_channels] regular convolutions, , nhwc data format.

import tensorflow tf import numpy np import random  width = 128 height = 128 channels = 32 kernel_width = 3 kernel_height = 3  tf.session() sess:      _input = np.float32(np.random.rand(1, height, width, channels))     _weights =  np.float32(np.random.rand(kernel_height, kernel_width, channels, 1))      _input_ph = tf.placeholder(tf.float32, shape=(1, height, width, channels))     _weights_pc = tf.placeholder(tf.float32, shape=(kernel_height, kernel_width, channels, 1))      feed = { _input_ph: _input, _weights_pc : _weights }      result = tf.nn.depthwise_conv2d(_input_ph, _weights_pc, [1,1,1,1], 'same')      individual_results = []     in range(channels):         individual_results.append(tf.nn.conv2d(tf.expand_dims(_input_ph[:,:,:,i],axis=3), tf.expand_dims(_weights_pc[:,:,i,:],axis=3), [1,1,1,1], 'same'))      depth_result = sess.run(result, feed_dict=feed)     concat_result = sess.run(tf.concat(individual_results, axis=3), feed_dict=feed)      channel_diff = 0.0     in range(channels):         channel_diff += np.sum(depth_result[:,:,:,i]-concat_result[:,:,:,i])      print(channel_diff) 

here i'm computing first normal tf.nn.depthwise_conv2d , slice input , weights accordingly , tf.nn.conv2ds individually. these parameters 1e-5 difference, tends higher when increase number of channels.

i glad if explain me what's going on :) thanks!