Snowflakes Removal for Single Image Based on Model Pruning and Generative Adversarial Network
Li He, Jie Zhang · 2019
The snow fluttering in the air seriously affects the quality of the image and destroys the completeness of the information in the picture. In this paper, we solve this problem by generative adversarial network training model and combining a method of model pruning. The neural network is used to remove snowflake. As the depth of the neural network increases, it is easy to appear over-fit, so we use model pruning optimization to reduce the computational volume of the deep learning model. In response to the above problems, we mainly deal with three steps: pruning the network; identify the snowflake area and use the area around the snowflake to recover the snowflake area; identify the restored picture to make sure that the restored image does not have much modification marks.