Optical Flow-Guided Mask Generation Network for Video Segmentation

The main theme of this project is to generate an optical flow-guided mask for video/image segmentation of animals. The main purpose of the segmentation is to differ the foreground objects.   In this project we propose CNN based technique for the semi-supervised video/image segmentation. Here we will train the data by using segmentation techniques for the segmentation of objects to get the ground truth values. 

Image/video will be treated as input to Unet architecture and it will create a mask over the segmented objects. By the use of Unet, we can easily create the mask over foreground segmented objects. The generated mask cannot be seen by using the Unet, hence we use semantic segmentation to visualize the segmented output. By using the semantic segmentation, we can visualize the foreground objects from the video/image.

Keywords: Semi-supervised, Video Object Segmentation, Optical Flow, Mask.