Dynamic-Net: Tuning the Objective Without Re-training for Synthesis Tasks
This addresses the inefficiency of re-training for synthesis tasks, offering a practical tool for users, though it is incremental as it builds on existing CNN optimization methods.
The paper tackles the problem of needing to re-train CNNs for different objectives by introducing Dynamic-Net, which allows tuning the objective at inference time without re-training, maintaining performance quality and enabling real-time interactive modifications.
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.