Video Rain Removal with D3R-Net Dynamic Routing Residue Recurrent Network

PROJECT TITLE :

D3R.Net Dynamic Routing Residue Recurrent Network for Video Rain Removal

ABSTRACT:

Rain occlusion regions, i.e., areas with low light transmittance for rain streaks, are addressed in this article. Unlike rain streaks that are added to the scene, the background is fully obliterated in these occlusion areas. Since rain streaks and occlusions are both present, we suggest a hybrid rain model. We introduce a Dynamic Routing Residue Recurrent Network (DRRN) that incorporates the hybrid model with valuable motion segmentation context data (D3R.Net). Using a residual network, D3R.Net first extracts the spatial features. The spatial features are then aggregated along the temporal axis using recurrent units. The context information is incorporated into the network in a 'dynamic routing' manner in the temporal fusion. Temporal fusion in certain scenarios, such as rain or non-rain regions is handled using a recurrent unit heap. Only one of these repeating components gets turned on in specific forward and reverse operations. Context-selecting gates then detect the context and pick one of these temporally fused features created by these recurrent units as the final fused feature. For the final time, this last characteristic serves as the residual feature. î Reconstructing the negative rain streaks requires combining it with a spatial feature. It's possible to build a D3R.Net that uses motion segmentation and rain type indicator as context variables to identify fast-moving and slow-moving edges, as well as occlusion and non-rain regions. Furthermore, extensive studies on a variety of synthetic and actual videos with rain streaks show that our network design and each component are both superior to and successful in comparison to current methods.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here