Abstract
Generative modeling has recently shown great promise in computer vision, but its success is often limited to separate tasks. In this paper, motivated by multi-task learning of shareable feature representations, we consider a novel problem of learning a shared generative model across various tasks. We instantiate it on the illustrative dual-task of joint few-shot recognition and novel-view synthesis: given only one or few images of a novel object from arbitrary views with only category annotation, we aim to simultaneously learn an object classifier and generate images of the object from new viewpoints. To this end, we propose bowtie networks that jointly learn 3D geometric and semantic representations with feedback in the loop. Experimental evaluation on challenging fine-grained recognition datasets demonstrates that our synthesized images are realistic from multiple viewpoints and significantly improve recognition performance as ways of data augmentation, especially in the low-data regime. We further show that our approach is flexible and can be easily extended to incorporate other tasks, such as style guided synthesis.
Abstract (translated)
URL
https://arxiv.org/abs/2008.06981
