Research on portability and effectiveness based on MIMO-Unet

In the field of computer vision, many excellent image restoration algorithms have emerged, and MIMO-Unet model has achieved a good effect in image deblurring. In this paper, MIMO-Unet model is taken as the research object to introduce the internal principle of MIMO-Unet, and the data set adopted is different from the open GOPRO data set collection means, and the self-made binocular camera is used to construct the data set. Among them, the blurry images and sharp images in the data set are captured by binocular camera. On this basis, the MIMO-Unet model is applied to deblur the images, so as to verify the effectiveness and portability of the MIMO-Unet model. In addition, compared with the public GOPRO data set, the image feature degradation of the data set made by binocular cameras is more serious, and the training effect of MIMO-Unet model on this data set is significantly reduced. In this paper, PSNR and SSIM are selected as two indicators of image defuzzing effect. The training results of the MIMO-Unet model on the binocular camera data set are 0.8028 (SSIM) and 28.87dB (PSNR), which are 0.15 and 6.93dB less than the training results on the GOPRO data set, respectively.