[经典网络]：Mask R-CNN 网络结构详解 : R-CNN, Fast R-CNN, Faster R-CNN, Mask R-CNN, FCN

论文：He, Kaiming, et al. “Mask r-cnn.” Proceedings of the IEEE international conference on computer vision. 2017.

代码：TensorFlow实现

实例分割模型Mask R-CNN详解：从R-CNN，Fast R-CNN，Faster R-CNN再到Mask R-CNN从R-CNN1开讲讲到Mask R-CNN2，很完整，简明的表述了R-CNN系列的发展，但是不够细节。Mask R-CNN2可以理解为Faster R-CNN3 + FCN4

1. R-CNN2算法详解

a. 在图像中确定约1000-2000个候选框

b. 对于每个候选框内图像块，使用深度网络提取特征

c. 对候选框中提取出的特征，使用分类器判别是否属于一个特定类

d. 对于属于某一特征的候选框，用回归器进一步调整其位置

1. Fast R-CNN5算法详解

网络末端同步训练的分类和位置调整，提升准确度

使用多尺度的图像金字塔，性能几乎没有提高

倍增训练数据，能够有2%-3%的准确度提升

网络直接输出各类概率(softmax)，比SVM分类器性能略好

更多候选窗不能提升性能

1. Faster R-CNN3算法详解

目标检测的四个基本步骤（候选区域生成，特征提取，分类，位置精修）终于被统一到一个深度网络框架之内

1. Mask R-CNN算法详解

   

   输入图片进行特征提取，使用FPN6网络

[5]:Lin, Tsung-Yi, et al. “Feature pyramid networks for object detection.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017.

参考：

1. Girshick, Ross, et al. “Rich feature hierarchies for accurate object detection and semantic segmentation.” Proceedings of the IEEE conference on computer vision and pattern recognition. 2014. ↩︎
2. He, Kaiming, et al. “Mask r-cnn.” Proceedings of the IEEE international conference on computer vision. 2017. ↩︎ ↩︎ ↩︎
3. Ren, Shaoqing, et al. “Faster r-cnn: Towards real-time object detection with region proposal networks.” Advances in neural information processing systems. 2015. ↩︎ ↩︎
4. Long, Jonathan, Evan Shelhamer, and Trevor Darrell. “Fully convolutional networks for semantic segmentation.” Proceedings of the IEEE conference on computer vision and pattern recognition. 2015. ↩︎
5. Girshick, Ross. “Fast r-cnn.” Proceedings of the IEEE international conference on computer vision. 2015. ↩︎
6. Lin, Tsung-Yi, et al. “Feature pyramid networks for object detection.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017. ↩︎