Image_Segmentation

基于深度学习的图像分割。图像分割包括语义分割、实例分割、全景分割。

语义分割模型

• FCN
• DeconvNet
• SegNet
• UNet
• PSPNet
• RefineNet
• GCN
• DeepLab（v1&v2&v3&v3+）
• PAN
• Auto-DeepLab
• NAS
• …

模型架构、代码复现等

语义分割论文

• Semantic Segmentation | Zhang Bin's Blog
• Segmentation - handong1587
• 语义分割 - Semantic Segmentation Papers - AIUAI

模型和复现

• mrgloom/awesome-semantic-segmentation
• guanfuchen/semseg - 常用的语义分割架构结构综述以及代码复现
• GeorgeSeif/Semantic-Segmentation-Suite - Semantic Segmentation Suite in TensorFlow. Implement, train, and test new Semantic Segmentation models easily!
• guanfuchen/DeepNetModel - 记录每一个常用的深度模型结构的特点（图和代码）
• handong1587.github.io/2015-10-09-segmentation.md

计算机视觉论文

• amusi/CVPR2019-Code: CVPR 2019 Paper with Code
• zziz/pwc: Papers with code. Sorted by stars
• amusi/daily-paper-computer-vision

代码实践

常看到代码中定义：

R = 103.939
G = 116.779
B = 123.68

什么意思？请看这里一个回答：https://gist.github.com/baraldilorenzo/07d7802847aaad0a35d3#gistcomment-1616734

Also, there is no normalization done in the gist above. If you want accurate results, you better do those steps to any input image:

    img = cv2.resize(cv2.imread('../../Downloads/cat2.jpg'), (224, 224))

     mean_pixel = [103.939, 116.779, 123.68]
     img = img.astype(np.float32, copy=False)
     for c in range(3):
         img[:, :, c] = img[:, :, c] - mean_pixel[c]
     img = img.transpose((2,0,1))
     img = np.expand_dims(img, axis=0)

The mean pixel values are taken from the VGG authors, which are the values computed from the training dataset.

最新：

• FDNet：学习全密集神经网络进行图像语义分割

  《Learning Fully Dense Neural Networks for Image Semantic Segmentation》(AAAI 2019)

  Date：2019 | Author：香港科技大学&微软亚洲研究院

  arXiv：https://128.84.21.199/abs/1905.08929

• 基于MobileNetV3的DeepLab V3+语义分割

  Mobile Deeplab-V3+ model for Segmentation

  This project is used for deploying people segmentation model to mobile device and learning. The people segmentation android project is here. The model is...

  arXiv：

• 遥感语义图像的边界损失

  《Boundary Loss for Remote Sensing Imagery Semantic Segmentation》

  Date:20190521 | Author: Aeronet

  arXiv：https://arxiv.org/abs/1905.07852

• HRNet：（告别低分辨率网络，微软提出高分辨率深度神经网络HRNet）

  《Deep High-Resolution Representation Learning for Human Pose Estimation》

  arXiv：https://arxiv.org/abs/1902.09212

  [1] Ke Sun, Bin Xiao, Dong Liu, Jingdong Wang: Deep High-Resolution Representation Learning for Human Pose Estimation. CVPR 2019

  [2] https://github.com/leoxiaobin/deep-high-resolution-net.pytorch

  [3] https://github.com/HRNet

  ——from：CVPR 2019 | 告别低分辨率网络，微软提出高分辨率深度神经网络HRNet

  《Deep High-Resolution Representation Learning for Human Pose Estimation》的原作者不仅把这种高分辨率网络结构用于姿态估计，也在尝试用于其他方向。

  不久前，作者在新论文《High-Resolution Representations for Labeling Pixels and Regions》中对网络结构进行了v2版本升级，给出了更多实验结果，更加验证了该网络结构的价值！

  在计算机视觉目前最热门应用领域语义分割、目标检测、人脸特征点定位中，换用高分辨率网络结构的算法都获得了显著的精度提升！

  arXiv：https://arxiv.org/abs/1904.04514

• 《Hard Pixels Mining: Learning Using Privileged Information for Semantic Segmentation》

  注：在NYU-v2数据集上，超越DeepLabV3+、PSPNet、RDFNet

  Date：2019 Author：上海交通大学

  arXiv：https://arxiv.org/abs/1906.11437