GitPlanet

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Created with love in Canada, visit hostnodejs.com today

Feel like to post an Ad? Learn Details
All Projects → LeslieZhoa → Tensorflow Deeplab_v3_plus

LeslieZhoa / Tensorflow Deeplab_v3_plus

图像分割算法deeplab_v3+，基于tensorflow，中文注释，摄像头可用

Programming Languages

python
139335 projects - #7 most used programming language

Labels

tensorflowvideoimage-segmentation

Projects that are alternatives of or similar to Tensorflow Deeplab v3 plus

Concise Ipython Notebooks For Deep Learning
Ipython Notebooks for solving problems like classification, segmentation, generation using latest Deep learning algorithms on different publicly available text and image data-sets.
Stars: ✭ 23 (-69.33%)
Mutual labels:  image-segmentation
Midv 500 Models
Model for document segmentation trained on the midv-500-models dataset.
Stars: ✭ 31 (-58.67%)
Mutual labels:  image-segmentation
Unet Tensorflow
Tensorflow implement of U-Net
Stars: ✭ 50 (-33.33%)
Mutual labels:  image-segmentation
Segmentation
Catalyst.Segmentation
Stars: ✭ 27 (-64%)
Mutual labels:  image-segmentation
Max Image Segmenter Web App
Deploy a Deep Learning Powered "Magic Cropping Tool" using Pre-Trained Open Source Models
Stars: ✭ 29 (-61.33%)
Mutual labels:  image-segmentation
Albumentations
Fast image augmentation library and an easy-to-use wrapper around other libraries. Documentation: https://albumentations.ai/docs/ Paper about the library: https://www.mdpi.com/2078-2489/11/2/125
Stars: ✭ 9,353 (+12370.67%)
Mutual labels:  image-segmentation
Eccv2020 Code
ECCV 2020 论文开源项目合集，同时欢迎各位大佬提交issue，分享ECCV 2020开源项目
Stars: ✭ 827 (+1002.67%)
Mutual labels:  image-segmentation
Multiclass Semantic Segmentation Camvid
Tensorflow 2 implementation of complete pipeline for multiclass image semantic segmentation using UNet, SegNet and FCN32 architectures on Cambridge-driving Labeled Video Database (CamVid) dataset.
Stars: ✭ 67 (-10.67%)
Mutual labels:  image-segmentation
Active Contour Model Python
The python code of Chan-Vese model and RSF model for image segmentation
Stars: ✭ 28 (-62.67%)
Mutual labels:  image-segmentation
Segmentationcpp
A c++ trainable semantic segmentation library based on libtorch (pytorch c++). Backbone: ResNet, ResNext. Architecture: FPN, U-Net, PAN, LinkNet, PSPNet, DeepLab-V3, DeepLab-V3+ by now.
Stars: ✭ 49 (-34.67%)
Mutual labels:  image-segmentation
Wbc segmentaion
White Blood Cell Image Segmentation Using Deep Convolution Neural Networks
Stars: ✭ 11 (-85.33%)
Mutual labels:  image-segmentation
Pytorch Toolbelt
PyTorch extensions for fast R&D prototyping and Kaggle farming
Stars: ✭ 942 (+1156%)
Mutual labels:  image-segmentation
Segmentation wbc
White blood cell (WBC) image datasets
Stars: ✭ 35 (-53.33%)
Mutual labels:  image-segmentation
Ternausnet
UNet model with VGG11 encoder pre-trained on Kaggle Carvana dataset
Stars: ✭ 928 (+1137.33%)
Mutual labels:  image-segmentation
Image Segmenter Android
Realtime Image Segmentation on Android
Stars: ✭ 55 (-26.67%)
Mutual labels:  image-segmentation
Ai sudoku
GUI based Smart Sudoku Solver that tries to extract a sudoku puzzle from a photo and solve it
Stars: ✭ 830 (+1006.67%)
Mutual labels:  image-segmentation
Graph Based Image Segmentation
Implementation of efficient graph-based image segmentation as proposed by Felzenswalb and Huttenlocher [1] that can be used to generate oversegmentations.
Stars: ✭ 31 (-58.67%)
Mutual labels:  image-segmentation
Lovaszsoftmax
Code for the Lovász-Softmax loss (CVPR 2018)
Stars: ✭ 1,148 (+1430.67%)
Mutual labels:  image-segmentation
Coco Annotator
✏️ Web-based image segmentation tool for object detection, localization, and keypoints
Stars: ✭ 1,138 (+1417.33%)
Mutual labels:  image-segmentation
Grabcutweb
Full web grabcut example using opencvjs
Stars: ✭ 40 (-46.67%)
Mutual labels:  image-segmentation
View All Similar Projects ➔

tensorflow-deeplab_v3_plus

参考rishizek的代码进行中文注释，并按照自己风格重新编写代码，对ASPP加入里BN层，支持摄像头。

deeplab_v3_plus简介

图像分割是主要功能是将输入图片的每个像素都分好类别，也相当于分类过程。举例来说就是将大小为[h,w,c]的图像输出成[h,w,1]，每个像素值代表一个类别。
deeplab_v3+可以参考论文Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation。它的结构图如下：


下面对模型进行简要分析
该模型属于encoder-decoder模型，encoder-decoder常用于自然语言处理中，在图像分割中U-net也是十分典型的encoder-decoder模型，大体结构如下：

就是将图片通过卷积尺寸变小再通过上采样将尺寸还原。

deeplab_v3+是将encoder-decoder和ASPP相结合，encoder-decoder会获取更多边界信息，ASPP可获取更多特征信息。encoder网络使用resnet101或 Xception,本代码中使用的是resnet101。

采用预训练的resnet的某一节点A来获取图像信息，再加入到ASPP中。ASPP就是不同膨胀率的空洞卷积和全局池化上采样后的输出concat在一起，作为encoder输出部分。

空洞卷积可以理解为一个大卷积中间权重值都为0,举例说明，一个3x3的卷积，如果膨胀率是1就是正常卷积，如果膨胀率是2,就是空洞卷积，相当于把3x3的卷积每个值的右方和下方加一行或列都置0。变换之后的空洞矩阵大小变为6x6。空洞矩阵论文中说可以提取更密集的特征，捕获多尺度信息，相比于卷积和池化会减少信息丢失。全局池化就是将输入[h,w,c]池化成[1,1,c]。

decoder部分选取resnet中A节点之前的B节点，再将encoder的输出上采样成B的大小然后concat，做一些卷积和上采样就得到最终输出。

由于可以看成分类问题，该模型的损失函数也是交叉熵函数。模型具体实现可以参考代码

模型训练

环境要求

ubuntu=16.04
tensorflow=1.4.1
opencv=3.4.1
windows下可以进行测试

下载数据集

将VOC解压到data目录下,下载DrSleep提供的SegmentationClassAug文件解压，如果访问不了可以尝试这个网址里面有_MACOSX和SegmentationClassAug两个文件夹，将里面的SegmentationClassAug放置到./data//VOCdevkit/VOC2012/下，里面的图片是shape为[h,w,1]每一个像素值都对应类别的label。

将restnet预训练数据解压放置在该模型的根目录下。

如果需要模型预训练数据可以将我训练的权重数据解压，如果下载过慢可以尝试这个网址,将里面的ckpt等文件放置到./model下。

代码介绍

data放置VOC数据和数据处理生成的record文件和分化数据的txt文件

model放置训练生成的模型和graph

output放置测试图片生成的分割图像

picture放置测试用例,我的来源于百度图片

utils包含配置文件config.py,数据处理文件dataset_util.py,preprocessing.py和模型文件deeplab_model.py

test.py是测试文件支持摄像头

tfrecord.py是将处理完的数据生成record文件

train.py是训练文件

运行

手动配置config.py的信息或选择默认

若要训练：
运行python tfrecord.py生成record文件
运行python train.py训练。 对于计算机gpu选择需修改这里代码

若要测试：
运行python test.py

一些疑问

我的电脑配置是1080Ti但总是运行运行就溢出，我尝试用tf.contrib.distribute.mirroredstrategy多gpu并行，但tensorflow版本要1.8,当我更新完，发现input_fn要是data格式，我失败了。
如果有并行gpu的建议或者代码的指正请给我留言

结果展示

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].