《Multi-Scale Continuous CRFs as Sequential Deep Networks for Monocular Depth Estimation》

这篇博文是深度估计（单目深度估计）

也是收到multi scale CNN 的启发，提出网络可以fuses complementary information derived from multiple CNN side outputs

the integration （融合） is obtained by means of continuous Conditional Random Fields (CRFs)

通过多尺度信息的融合，实现单目深度估计

fusing feature maps corresponding to different network layers or designing an architecture with multiple inputs corresponding to images at different resolutions.

by adding a Conditional Random Field (CRF) in cascade to a convolutional neural architecture, the performance can be greatly enhanced and the CRF can be fully integrated within the deep model enabling end-to-end training with back-propagation

 

contributions:

1、提出了新方法从RGB图中来预测深度图，利用CNN内层产生的多尺度图并通过CRF framework来融合。

2、由于pixel 水平的深度预测意味着需要推断一系列的连续值。作者展示了如何将平均场（MF）更新作为顺序深层模型来实现，从而实现整个网络的端到端的训练（mean field（MF））。

3、our experiments demonstrate that the proposed multi-scale CRF framework is superior to previous methods integrating information from intermediate network layers by combining multiple losses  or by adopting feature concatenation （我们的实验表明，所提出的多尺度CRF框架优于以前通过组合多个损失或采用特征连接来集成中间网络层信息的方法。）

接下来看网络结构~

Multi-Scale Models for Depth Estimation

网络结构如下图所示