More explanation regarding fusion module

[AdityaNikhil]

Hi there,

I am working on re-implementing your paper in TensorFlow. The details you provided regarding Instance and full image networks are straight-forward. But I request you to please provide more instructions on how exactly to fuse these to get results. Especially I am confused about how to utilize the bounding box co-ordinates of each instance and marking the location inside the full image and training on it.

[lkp520]

I also need more details of fusion module.