Consolidated Adversarial Network for Video De-raining and De-hazing

The performance of recent video enhancement methods is superior in specific hazy, rainy, snowy, and foggy weather conditions. However, these approaches can handle degradation rendered by single weather. We propose an integrated lightweight adversarial learning network to handle the degradations induced by different weather conditions. This is a unique approach to mitigate the problem of video restoration for multi- weather degraded videos using single network. The proposed architecture combines the idea of multi-resolution analysis with a multi-scale encoder and domain-specific feature learning is achieved using domain-aware filtering modules. The architecture provides recurrent feature sharing for temporal consistency, achieved by feeding the previous frame output as feedback. Substantial experiments on various datasets demonstrate that the proposed method performs competitively with the existing state-of-the-art approaches for video restoration in multi-weather conditions.