当前位置:网站首页>Image classification, object detection, semantic segmentation, instance segmentation and panoramic segmentation

Image classification, object detection, semantic segmentation, instance segmentation and panoramic segmentation

2021-06-23 23:45:27 Wangqi

1、Image Classification( Image classification )
Image classification ( The following figure on the left ) Is to determine the classification of the image , For example, in learning classification, there are people in the data set (person)、 sheep (sheep)、 Dog (dog) And the cat (cat) Four kinds of , Image classification requires a given image output image contains which classification , For example, the example below contains person、sheep and dog Three .

2、Object detection( object detection )
object detection ( Top right ) In short, what's in the picture ? Where are the differences ?( Frame them in rectangles )

At present, the commonly used target detection algorithms are Faster R-CNN And based on YOLO The algorithm of target detection based on

3、semantic segmentation( Semantic segmentation )
Usually, target segmentation refers to semantic segmentation

Semantic segmentation ( The following figure on the left ) You need to distinguish every pixel in the picture , It's not just the rectangle . But different instances of the same object don't need to be separated . To the left of the picture below , Mark as a person , sheep , Dog , The grass . And you don't need sheep 1, sheep 2, sheep 3, sheep 4, sheep 5 etc. .

4、Instance segmentation( Instance segmentation )
Instance segmentation ( Top right ) In fact, it is the combination of target detection and semantic segmentation . Relative to the bounding box of target detection , Instance segmentation can be accurate to the edge of the object ; Relative semantic segmentation , Instance segmentation needs to mark different individuals of the same object on the graph ( sheep 1, sheep 2, sheep 3...)

At present, the commonly used instance segmentation algorithm is Mask R-CNN.

Mask R-CNN Through to the Faster R-CNN Add a branch for pixel level segmentation , This branch outputs a binary mask , This mask indicates whether a given pixel is part of the target object : This branch is a full convolution network based on convolution neural network feature mapping . The given convolution neural network feature map is used as input , The output is a matrix , Where pixels belong to all the positions of the object 1 Express , For other locations, use 0 Express , This is the binary mask .

Once these masks are generated , Mask R-CNN take RoIAlign And from Faster R-CNN A combination of classification and bounding box , For accurate segmentation :

5、Panoramic segmentation( Panoramic segmentation )
Panoramic segmentation is a combination of semantic segmentation and instance segmentation . Unlike instance segmentation : Instance segmentation only applies to object To test , And to detect object Segmentation , Panoramic segmentation is to detect and segment all the objects in the image, including the background .


————————————————

Link to the original text :https://blog.csdn.net/kk123k/article/details/86584216

版权声明
本文为[Wangqi]所创,转载请带上原文链接,感谢
https://chowdera.com/2021/06/20210623234449584V.html

随机推荐