Compressive Convolutional Network (CCN)

Intruduction

The proposed algorithm of the Compressive Convolutional Network (CCN) features the ability to perform near-isometric compressive sensing using convolutional operations. A novel incoherent convolution approach is invented for learning the embedded matrix to achieve near-isometric property for compressive sensing.

Dependencies

• Matlab
• C++

Run Compressive Convolutional Network

FrontEnd_Detection&Compression is the training code, which is modified from the official public YOLOv2 code. demo.m is a demo to compress and reconstruct images with different compression ratios. You can run demo.m to gain the reconstruction image with evaluation indexes of PSNR and SSIM.

Experiments

Dataset	BSD100	BSD100	VOC	VOC
Methods	PSNR	SSIM	PSNR	SSIM
CCN-YOLO	26.56	0.8192	26.54	0.8786
CCN-SSD	25.19	0.7872	25.46	0.8253
ADAGIO	22.42	0.6055	23.89	0.6325
RandConv	22.31	0.6243	22.21	0.6608
CS-SM	21.39	0.5954	21.46	0.6217
GAUSS	21.32	0.5921	22.48	0.6409