Original Input image from three POV, which will be masked by depth data. Then OpenCV filtering.

origin author PoKong Lai

Visualizating the 3D output to ensure the Camera Poses was properly predicted.

This Visualizes the environment with three relative camera locations.

The major challenges is how to synchronizes these camera. So we uses SLAM and VR point-cloud scattering to predict the poses.

Recording a person from three point of views. And perform loop closure to predict the camera poses of these POV.

• 
  • This is the initial mask that will be selected. For each image we will compute similarity to pick the most same components as filter.

network architecture

link for result in prediction of backward image from front image.

• left most column is input, middle is ground truth and last is prediction.
• feed front image hence left most to AutoEncoder to generate backward image.