Depthwise Separable Convolution
A convolution variant that decomposes standard convolution into two steps, depthwise convolution and pointwise convolution, substantially reducing computation and parameter count. It is the core design of lightweight networks such as MobileNet.
Timeline (last 90 days)
Depthwise Separable Convolution is the core building block proposed by the MobileNet series; it splits standard convolution into two steps, depthwise and pointwise, and under 3×3 convolution it can usually reduce computation to one-eighth to one-ninth of the original.
Depthwise separable convolution decomposes standard convolution into two steps, depthwise convolution and pointwise convolution, reducing computation and parameter count by about 8 to 9 times, at the cost of delaying inter-channel information interaction and slightly losing representational capacity.
The frame decoder uses 4-level bilinear upsampling combined with depthwise separable convolutions to decode a 64-dimensional latent vector into a 384×512 grayscale image.
Depthwise separable convolutions reduce the parameter count of standard convolutions from K×K×C_in×C_out to K×K×C_in + C_in×C_out, typically achieving an 8–9× reduction.
All Facts (4)
Depthwise Separable Convolution is the core building block proposed by the MobileNet series; it splits standard convolution into two steps, depthwise and pointwise, and under 3×3 convolution it can usually reduce computation to one-eighth to one-ninth of the original.
50%UnverifiedDepthwise separable convolution decomposes standard convolution into two steps, depthwise convolution and pointwise convolution, reducing computation and parameter count by about 8 to 9 times, at the cost of delaying inter-channel information interaction and slightly losing representational capacity.
50%UnverifiedThe frame decoder uses 4-level bilinear upsampling combined with depthwise separable convolutions to decode a 64-dimensional latent vector into a 384×512 grayscale image.
50%UnverifiedDepthwise separable convolutions reduce the parameter count of standard convolutions from K×K×C_in×C_out to K×K×C_in + C_in×C_out, typically achieving an 8–9× reduction.
50%