Alon Shoshan, Roey Mechrez, Lihi Zelnik-Manor
Technion - Israel Institute of Technology

Abstract:
One of the key ingredients for successful optimization of modern CNNs is identifying a suitable objective. To date, the objective is fixed a-priori at training time, and any variation to it requires re-training a new network. In this paper we present a first attempt at alleviating the need for re-training. Rather than fixing the network at training time, we train a "Dynamic-Net" that can be modified at inference time. Our approach considers an "objective-space" as the space of all linear combinations of two objectives, and the Dynamic-Net is emulating the traversing of this objective-space at test-time, without any further training. We show that this upgrades pre-trained networks by providing an out-of-learning extension, while maintaining the performance quality. The solution we propose is fast and allows a user to interactively modify the network, in real-time, in order to obtain the result he/she desires. We show the benefits of such an approach via several different applications.

Control over Stylization level

Interpolation between two Styles

The proposed method allow us to generate faces with control over the facial attributes e.g gender or hair color

Dynamic-Net allows the user to select the best working point for each image, improving results of networks that were trained with sub-optimal objectives

Code for every application is written as a separate project:

• Dynamic Style Transfer
• Dynamic DCGAN

Dynamic style transfer demo

If you use our code for research, please cite our paper:

@InProceedings{Shoshan_2019_ICCV,
  title = {Dynamic-Net: Tuning the Objective Without Re-Training for Synthesis Tasks},
  author = {Shoshan, Alon and Mechrez, Roey and Zelnik-Manor, Lihi},
  booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
  month = {October},
  year = {2019}
}

Code for the style transfer network implementation borrows from [1][2].
Code for the DCGAN network implementation borrows from [1].