FCN

PaperReading TextDetection 2017.12

[1] Fully Convolutional Networks for semantic Segmentation

Abstract

1. Introduction

2. Related Work

3. Fully convolutional networks

• Adapting classfier for dense prediction

• Shift-and-stitch is filter rarefaction

  最后的输出是downsampling的，如何做调整以实现semantic segmentation?
  trick:按行填0，使得最后卷积的输出是输入的receptive field的中心点。

• Unsampliing is backwards srided convolution

  还有一种方法是插值。

  In a sense, upsampling with factor $f$ is convolution with a fractional input stride of $1/f$ . So long as $f$ is integral, a natural way to upsample is therefore backwards convolution (sometimes called deconvolution) with an output stride of $f$ .

  是transposed convolution

• Patchwise training is loss sampling

4. Segmentation Architecture

• Fom classifier to dense FCN

• Combing what and where

• Experimental framwork

5. Notes

所以说这篇论文的方法就是先通通过前向的卷积产生一些feature map，然后在此基础上用deconvolution的方法upsample，最后完成pixelwise的分类吗？