**Synchronized Batch Normalization (SyncBN)** is a type of [batch normalization](https://paperswithcode.com/method/batch-normalization) used for multi-GPU training. Standard batch normalization only normalizes the data within each device (GPU). SyncBN normalizes the input within the whole mini-batch.

A **Dimension-wise Convolution**, or **DimConv**, is a type of [convolution](https://paperswithcode.com/method/convolution) that can encode depth-wise, width-wise, and height-wise information independently. To achieve this, DimConv extends depthwise convolutions to all dimensions of the input tensor $X \in \mathbb{R}^{D\times{H}\times{W}}$, where $W$, $H$, and $D$ corresponds to width, height, and depth of $X$. DimConv has three branches, one branch per dimension. These branches apply $D$ depth-wise convolutional kernels $k\_{D} \in \mathbb{R}^{1\times{n}\times{n}}$ along depth, $W$ width-wise convolutional kernels $k\_{W} \in \mathbb{R}^{n\times{1}\times{1}}$ along width, and $H$ height-wise convolutional kernels $k\_{H} \in \mathbb{R}^{n\times{1}\times{n}}$ kernels along height
to produce outputs $Y\_{D}$, $Y\_{W}$, and $Y\_{H} \in \mathbb{R}^{D\times{H}\times{W}}$ that
encode information from all dimensions of the input tensor. The outputs of these independent branches are concatenated along the depth dimension, such that the first spatial plane of $Y\_{D}$, $Y\_{W}$, and $Y\_{H}$ are put together and so on, to produce the output $Y\_{Dim} = ${$Y\_{D}$, $Y\_{W}$, $Y\_{H}$} $\in \mathbb{R}^{3D\times{H}\times{W}}$.

DimConv

DiCENet: Dimension-wise Convolutions for Efficient Networks

SyncBN

Context Encoding for Semantic Segmentation

**Libra R-CNN** is an object detection model that seeks to achieve a balanced training procedure. The authors motivation is that training in past detectors has suffered from imbalance during the training process, which generally consists in three levels – sample level, feature level, and objective level. To mitigate the adverse effects, Libra R-CNN integrates three novel components: IoU-balanced
sampling, [balanced feature pyramid](https://paperswithcode.com/method/balanced-feature-pyramid), and [balanced L1 loss](https://paperswithcode.com/method/balanced-l1-loss), respectively for reducing the imbalance at sample, feature, and objective level.

Source	Context Encoding for Semantic Segmentation
Year	2000
Data Source	CC BY-SA - https://paperswithcode.com

Viet-Anh on Software

What is: Synchronized Batch Normalization?

Viet-Anh on Software