Switchable Normalization for Learning-to-Normalize Deep Representation

Abstract

We address a learning-to-normalize problem by proposing Switchable Normalization (SN), which learns to select different normalizers for different normalization layers of a deep neural network. SN employs three distinct scopes to compute statistics (means and variances) including a channel, a layer, and a minibatch. SN switches between them by learning their importance weights in an end-to-end manner. It has several good properties. First, it adapts to various network architectures and tasks (see Fig. 1). Second, it is robust to a wide range of batch sizes, maintaining high performance even when small minibatch is presented (e.g. 2 images/GPU). Third, SN does not have sensitive hyper-parameter, unlike group normalization that searches the number of groups as a hyper-parameter. Without bells and whistles, SN outperforms its counterparts on various challenging benchmarks, such as ImageNet, COCO, CityScapes, ADE20K, MegaFace and Kinetics. Analyses of SN are also presented to answer the following three questions: (a) Is it useful to allow each normalization layer to select its own normalizer? (b) What impacts the choices of normalizers? (c) Do different tasks and datasets prefer different normalizers? We hope SN will help ease the usage and understand the normalization techniques in deep learning. The code of SN has been released at https://github.com/switchablenorms.

Differentiable Learning-to-Group Channels via Groupable Convolutional Neural Networks

“…(6). The optimization of binary variables has been well established in the literature [22,18,17,26], which can be also used to train DGConv. The gate params are optimized by Straight-Through Estimator similar to recent network quantization approaches, which is guaranteed to converge [5].…”

Stable and Effective One-Step Method for Person Search

“…In this case, BN is prone to introduce random noise during the training process. To solve this problem, we firstly introduce Switchable normalization (SN) [18] to person search, which is proposed to learn different operations in different normalization layers of a deep neural network. The candidate normalizer of SN includes Instance Normalization (IN) [23], Layer Normalization (LN) [24] and Batch Normalization (BN) [22], which estimates channel-wise, layer-wise and minibatch-wise statistics separately.…”

“…ω in (k ∈ {in, ln, bn}) are important weights learned by SN during training. More details refer to [18].…”

UCT‐GAN: underwater image colour transfer generative adversarial network

“…Therefore, we combined the up‐sampling layer and the convolutional block to realise up‐sampling. Each convolutional block architecture uses the same combination of a convolutional layer and switchable normalisation [27] followed by the leaky rectified linear units activation layer. SN is the improved combination of batch normalisation, instance normalisation and layer normalisation, and it can learn to select different normalisers for different normalisation layers of a deep neural network.…”