Skip to main content

Table 5 Single-inference accuracies of networks with individually quantized network parts

From: A depthwise separable convolutional neural network for keyword spotting on an embedded system

Floating point accuracy 83.2 % (79.3 %)
Bit width8-bit4-bit2-bit
Convolution weights83.2 % (79.1 %)82.3 % (78.8 %)46.8 % (44.6 %)
DW-Convolution weigths83.3 % (79.3 %)80.0 % (76.7 %)12.0 % (12.1 %)
PW-Convolution weights83.4 % (79.2 %)78.9 % (75.4 %)11.2 % (10.8 %)
Fully connected weights83.2 % (79.3 %)82.9 % (79.1 %)70.1 % (63.9 %)
Layer activations83.2 % (79.2 %)53.6 % (52.0 %)8.6 % (8.7 %)