Brain and cognitive neuroscience matlab psychoolbox cognitive science experimental design - experimental design 4

Matlab Psytoolbox（ Brain and cognitive neuroscience ）

Examples of optical illusion
Experimental design 1
Experimental design II
Experimental design 3
Experiment design IV

---

Preface

Brain and cognitive neuroscience Matlab Cognitive science significance test （ Attention prediction ）

1. subject

function MIT/Tuebingen Saliency Benchmark（ Website ：https://saliency.tuebingen.ai/） The code on , Find your own pictures , Show run results , Know about significance detection

2. Screenshot

After entering the website , Click on results, There are many models to choose from

I chose RARE2012 and Context-Aware Saliency Two models , As a comparison of attention prediction . Click on the name , Go to the link , The two interfaces are
RARE2012, Click on the red area below to download the thesis and source code
Context-Aware Saliency, The interface also has the following links to papers and download source code , Click to

The original picture when I execute the code is these six

The running results are respectively
RARE2012

Context-Aware Saliency

Both are matlab function ,RARE2020 The execution time is approximately 10s within , The result is rough , and Context-Aware Saliency The execution time varies from tens of seconds , The predicted area is more in line with human perception .

3. Evaluate the results

Because the webpage provides python There are many evaluation codes bug, I don't recommend using that evaluation code , You can go to github Find your own evaluation code . Or directly use the existing evaluation data .

The meanings of different indicators are explained in detail below ：
AUC： This is a ROC A version of the area under the curve . Saliency mapping is used as a binary classifier to distinguish positive samples and negative samples under different thresholds .
sAUC：Shuffled AUC Is under the area ROC A version of curve measurement . Saliency mapping is used as a binary classifier to distinguish positive samples and negative samples under different thresholds .
NSS：The normalized scanpath saliency, Measure the normalized significance diagram ( Zero mean , Unit variance ) Average significance value of fixed position .
CC：The correlation coefficient, Is the linear correlation coefficient between the model significance diagram and the empirical significance diagram , The saliency map is obtained by convolution of gaze position and Gaussian kernel .
KLDiv：Kullback-Leibler divergence, Normalize the model significance diagram and empirical significance diagram , Divide the sum by the density , Then calculate the Kullback-Leibler The divergence .
SIM：Similarity, This similarity measure is also called histogram intersection , When considered as a distribution , It measures the similarity between two different saliency graphs . Its calculation method is , Firstly, the significance mapping and empirical significance mapping of the model are normalized , Divide it by the sum , Then add the pixel level minimum values of the two distributions .

RARE2012

Context-Aware Saliency

4. Code changes

Because the source code cannot execute multiple pictures at one time , I make some modifications , Make it possible to display the results of six pictures at one time
（ Pay attention to modifying the picture name ）

RARE2012 Of example.m Executable files

I=cell(6);
R=cell(6);
I{
    1} = im2double(imread('images/1.jpeg'));
I{
    2} = im2double(imread('images/2.jpeg'));
I{
    3} = im2double(imread('images/3.jpeg'));
I{
    4} = im2double(imread('images/4.jpeg'));
I{
    5} = im2double(imread('images/5.jpeg'));
I{
    6} = im2double(imread('images/6.jpeg'));

R{
    1} = Rare2007(I{
    1});
R{
    2} = Rare2007(I{
    2});
R{
    3} = Rare2007(I{
    3});
R{
    4} = Rare2007(I{
    4});
R{
    5} = Rare2007(I{
    5});
R{
    6} = Rare2007(I{
    6});
N = length(R);

for i=1:N
    figure(i); clf;
    subplot(1,2,1); imshow(I{
    i});  title('Initial image')
    subplot(1,2,2); imshow(R{
    i}); title('Raw saliency map')
end    

Context-Aware Saliency Of run_saliency.m Executable files

%% A script for running saliency computation
clear all;
close all;

%% load parameters and images


file_names{
    1} = '1.jpeg';
file_names{
    2} = '2.jpeg';
file_names{
    3} = '3.jpeg';
file_names{
    4} = '4.jpeg';
file_names{
    5} = '5.jpeg';
file_names{
    6} = '6.jpeg';
MOV = saliency(file_names);


%% display results
N = length(MOV);
for i=1:N
    figure(i); clf;
    subplot(1,2,1); imshow(MOV{
    i}.Irgb); title('Input','fontsize',16);
    subplot(1,2,2); imshow(MOV{
    i}.SaliencyMap); title('Saliency map','fontsize',16);
end   

summary

Through the comparison of method 2 and method 1 , We can visually feel that the model of method 2 is better than method 1 , The images displayed after running the code of methods 1 and 2 have edge detection algorithms , Method 1 starts with the outline of things , Emphasize the main part , But there are too many white parts , Unable to focus on which part . Relatively speaking, the focus area can be clearly seen in the image after executing the code of method 2 , Some are ignored , Focus on the whole , Prevent the main part from local interference , Prevent important areas from being distorted , The marked area is closer to the attention area . Of course , Method 2 also takes longer to execute the code .
Method 2 running time ： Average time for four of the six pictures 30s about , Only figure 2 and figure 3 are exceptions , After observation and analysis , Figure 2 and figure 3 are animated pictures , colorful , Too much halo , Personally, I think these aspects lead to a long code execution time , Figure 2 40-50s, Figure 3 time consuming 75s about .
Method 1 running time ：10s The running results of six pictures are displayed in .

Thank you very much for watching , The content has been uploaded completely , In addition, my home page also has python Artificial intelligence , SCM and other articles , Welcome to