Learning Implicit Generative Models by Matching Perceptual Features
By C. D. Santos, Youssef Mroueh, Inkit Padhi, Pierre L. Dognin. 2019
32 citations recorded by the source corpus.
Perceptual features (PFs) have been used with great success in tasks such as transfer learning, style transfer, and super-resolution. However, the efficacy of PFs as key source of information for learning generative models is not well studied. We investigate here the use of PFs in the context of learning implicit generative models through moment matching (MM). More specifically, we propose a new effective MM approach that learns implicit generative models by performing mean and covariance matching of features extracted from pretrained ConvNets. Our proposed approach improves upon existing MM methods by: (1) breaking away from the problematic min/max game of adversarial learning; (2) avoiding online learning of kernel functions; and (3) being efficient with respect to both number of used moments and required minibatch size. Our experimental results demonstrate that, due to the expressiveness of PFs from pretrained deep ConvNets, our method achieves state-of-the-art results for challenging benchmarks.
Subfields: Computer Vision and Pattern Recognition, Generative Models, Signal Processing
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Quoc V. Le
- Andreas Geiger
- Chunhua Shen
- Ming-Hsuan Yang
- Jiashi Feng
- Xiaogang Wang
In Generative Models
- Axel Sauer
- Andreas Geiger
- Mark Sandler
- Jens Muller
- Kashyap Chitta
- Andrey Zhmoginov
- Tero Karras
- Timo Aila
In Signal Processing
Citations in the corpus
- Projected GANs Converge Faster (2021, background): “The non-adversarial approach of [53] learns generative models with moment matching in pretrained models; however, the results remain far from competitive to standard GANs.”
- Large-Scale Generative Data-Free Distillation (2020, methodology): “Therefore, we can force the layer statistics (mean and variance specifically) produced by our synthetic images to align with those emerging from the real data [10, 48, 53].”