20220524 deep learning technology points

1、1*1 Convolution nuclear action , Dimension reduction
1*1 The function of convolution kernel _nefetaria The blog of -CSDN Blog _1*1 The function of convolution kernel
2、 Image semantic information
For the semantic information in the image 、 Understanding of high-level and low-level features _Brucechows The blog of -CSDN Blog _ Low level features and high-level features

FPN Solve small object detection , High semantic  

3、select search 

Selective Search ( Select Search )_JNingWei The blog of -CSDN Blog _search selective

4、 Advantages of convolution kernel

Advantages of convolution kernel - Weight sharing and local perception _ Yenan liu The blog of -CSDN Blog _ Convolution layer local perception
5、 Neuron 、 Number of connections per neuron

What is in convolutional neural networks -----“ Neuron ” as well as “ The number of connections ”_ct Yuda's blog -CSDN Blog _ Convolutional neural network neuron

6、ROI pooling. Feature pooling

ROI POOLING Introduce - Good big pig - Blog Garden
7、 Over fitting reason
The network layer is very deep and has many parameters , Too little data , Easy to produce

8、 Normalized difference
Input normalization 、 Batch normalization (BN) Normalized with layer (LN)_ Xuan yanru Liu's blog -CSDN Blog _ Layer normalization

9、ocr Degree of confidence

video OCR Text confidence threshold in （threshold） What is the function of parameters ？_ Video analysis services VAS_ common problem _ Hua Wei Yun

10、 Deep learning head、neck、backbone What do the three terms mean ？

Deep learning head、neck、backbone What do the three terms mean ？_AI The world's leading blog -CSDN Blog _ Deep learning head

11、yolo Loss function

http://www.136.la/jingpin/show-141192.html

12、 Mask in image (Mask) What is it? _bitcarmanlee The blog of -CSDN Blog _ A mask  

13、 Face recognition

Generally speaking , There are three steps to face recognition ：

1. Find face ： Find the area with face in the picture and frame it
2. Align faces ： Mark the glasses, nose, mouth, etc. of the face , Use this as a basis to align faces
3. distinguish ： Recognize the aligned face , Determine who this face is
    

14、 Measure learning

Metric Learning—— Measure learning - Always add oil to yourself - Blog Garden

15、 Image flooding 、 Connected domain

Mask in image (Mask) What is it? _bitcarmanlee The blog of -CSDN Blog _ A mask

OpenCV—— Image connected domain _ I have a dream blog -CSDN Blog _opencv Connected domain

16、sift Image matching

SIFT Image matching _Yep_Ying The blog of -CSDN Blog _sift Image matching

17、 Principle of marking

Principle of target detection and marking _..Severus The blog of -CSDN Blog _ Target detection and labeling