LucidDream: Controlled Temporally-Consistent DeepDream on Videos
This work addresses video editing challenges for creators using neural network-based hallucination techniques, but it is incremental as it builds on existing DeepDream methods.
The paper tackled the problem of applying DeepDream to videos by improving controllability over hallucinated object classes and reducing flickering artifacts, achieving enhanced aesthetic appeal through a simple modification and a temporal consistency loss.
In this work, we aim to propose a set of techniques to improve the controllability and aesthetic appeal when DeepDream, which uses a pre-trained neural network to modify images by hallucinating objects into them, is applied to videos. In particular, we demonstrate a simple modification that improves control over the class of object that DeepDream is induced to hallucinate. We also show that the flickering artifacts which frequently appear when DeepDream is applied on videos can be mitigated by the use of an additional temporal consistency loss term.