This project is to build a lightweight and high efficiency CNN models，it is constructed by the core convolutional block unit named DSAC(Dual-scale separsable attention convolution). The DSAC unit is designed by separating the CBAM(convolution block attention module) and embedding CBAM into the depthwise separsable convolution. The DSAC has two different scale convolution kenel size which is 3×3 and 5×5, and it’s architecture is DW-SAM-PW-CAM. The DSACNet has novel architecture dual scale DW-SAM-PW-CAM，which also named DSAC.
We construct a DSACNet by only using 5 DSAC unit，and it has 1.02M parameters and 24.18MFlops
Please contact me in [email protected] if you want more and complete project