语义分割漫漫长路

本文主要是介绍语义分割漫漫长路，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！

一些资料小记录

Philipp Krähenbühl

德克萨斯大学奥斯汀分校计算机科学系的助理教授。
https://www.philkr.net/
主页有论文和代码
这里写图片描述

2017
Sampling Matters in Deep Embedding Learning
2016
Generative Visual Manipulationon the Natural Image Manifold
…
2011
Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials
关键字：DenseCRFs

MX Mask R-CNN

https://github.com/dragonfly90/mask-product

这里写图片描述

Piecewise CRF

This is an implementation of piecewise crf training for semantic segmentation based on the work of Chen et al. The
implemented model consists of three parts:

A neural network used for learning unary and binary potentials
A contextual conditional random field that combines the learnt unary and binary potentials
A fully connected Gaussian conditional random field used for segmentation postprocessing

这里写图片描述
https://github.com/Vaan5/piecewisecrf

cnn-densecrf-kitti-public

===== Description =====This software pertains to the research described in the CVPR 2016 paper:
Instance-Level Segmentation for Autonomous Driving with Deep Densely Connected MRFs
Ziyu Zhang, Sanja Fidler and Raquel UrtasunIn annotations/, we make our manual car annotations publicly available.In cnn/, we provide trained deeplab model for segmenting and depth
ordering instances in local image patches.In densecrf/, we provide the CRF framework we use for merging patches.In image_sets/, we list the training/validation/test split we use in
our paper. All images in three sets come from the training set of
KITTI's object detection benchmark.

https://bitbucket.org/zhangziyu1991/cnn-densecrf-kitti-public

这里写图片描述
Instance-Level Segmentation for Autonomous Driving with Deep Densely Connected MRFs

libDenseCRF

DescriptionThis is a modified version of a forked densecrf, which was used as a part of the DeepLab.

https://github.com/cvlab-epfl/densecrf

PyDenseCRF

https://github.com/lucasb-eyer/pydensecrf

DenseCRF

CodeThe code is modified from the publicly available code by Philipp Krähenbühl and VladlenKoltun. See their project website for more informationIf you also use this part of code, please cite their paper: Efficient Inference in Fully 
Connected CRFs with Gaussian Edge Potentials, Philipp Krähenbühl and Vladlen Koltun, NIPS 2011.