arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1610.04490cs.CVcs.LGstat.ML
收藏

图像超分辨率的摊销MAP推断

Amortised MAP Inference for Image Super-resolution

University of Copenhagen(哥本哈根大学);Twitter

Casper Kaae Sønderby, Jose Caballero, Lucas Theis, Wenzhe Shi, Ferenc Huszár

摘要:

图像超分辨率(SR)是一个欠定逆问题,其中大量看似合理的高分辨率图像可以解释相同的下采样图像。大多数当前的单图像SR方法使用经验风险最小化,通常具有像素均方误差(MSE)损失。然而,这种方法的输出往往是模糊的,过度平滑的,通常看起来难以置信。一个更理想的方法将采用最大后验(MAP)推理,更喜欢的解决方案,总是有一个高概率下的图像之前,因此似乎更合理。SR的直接MAP估计是不平凡的,因为它需要我们从样本中建立图像先验模型。此外,MAP推理通常是通过基于优化的迭代算法来执行的,这与基于神经网络的替代方案的效率相比并不好。在这里,我们介绍新的方法摊销MAP推理,我们直接使用卷积神经网络计算MAP估计。我们首先介绍了一种新的神经网络架构,该架构执行有效SR解决方案的仿射子空间的投影,确保网络的高分辨率输出始终与低分辨率输入一致。我们表明,使用这种架构,摊销MAP推理问题减少到最小化两个分布之间的交叉熵,类似于训练生成模型。我们提出了三种方法来解决这个优化问题:(1)生成对抗网络(GAN)(2)去噪引导SR,它将去噪的梯度估计反向传播到训练网络,以及(3)使用最大似然训练图像先验的基线方法。我们的实验表明,基于GAN的方法在真实图像数据上表现最好。最后,我们建立了GANs和摊销变分推理之间的联系,例如变分自编码器。

英文摘要:

Image super-resolution (SR) is an underdetermined inverse problem, where a large number of plausible high-resolution images can explain the same downsampled image. Most current single image SR methods use empirical risk minimisation, often with a pixel-wise mean squared error (MSE) loss. However, the outputs from such methods tend to be blurry, over-smoothed and generally appear implausible. A more desirable approach would employ Maximum a Posteriori (MAP) inference, preferring solutions that always have a high probability under the image prior, and thus appear more plausible. Direct MAP estimation for SR is non-trivial, as it requires us to build a model for the image prior from samples. Furthermore, MAP inference is often performed via optimisation-based iterative algorithms which don't compare well with the efficiency of neural-network-based alternatives. Here we introduce new methods for amortised MAP inference whereby we calculate the MAP estimate directly using a convolutional neural network. We first introduce a novel neural network architecture that performs a projection to the affine subspace of valid SR solutions ensuring that the high resolution output of the network is always consistent with the low resolution input. We show that, using this architecture, the amortised MAP inference problem reduces to minimising the cross-entropy between two distributions, similar to training generative models. We propose three methods to solve this optimisation problem: (1) Generative Adversarial Networks (GAN) (2) denoiser-guided SR which backpropagates gradient-estimates from denoising to train the network, and (3) a baseline method using a maximum-likelihood-trained image prior. Our experiments show that the GAN based approach performs best on real image data. Lastly, we establish a connection between GANs and amortised variational inference as in e.g. variational autoencoders.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑