-
Notifications
You must be signed in to change notification settings - Fork 0
/
Run_PFI.m
191 lines (160 loc) · 6.92 KB
/
Run_PFI.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
% Result 2b: Viewpoint-invariant unit and specific units and its visual feature encoding
%% Perform PFI analysis or load saved result
filename = strcat('./Result/RESULT_PFI_', target_class, '_', var_type, '.mat');
if exist(filename) == 0
disp(['Perform PFI analysis ...'])
PFI_RESULT = cell(4, 1);
% Set target units to analyze PFI
target_IND_Cell = cell(4, NN);
for nn=1:NN
target_IND_Cell{1,nn} = Cell_Inv{5,1,nn}; % invariant units
target_IND_Cell{2,nn} = Cell_Inv{5,2,nn}(Cell_Inv{5,3,nn}==1); % specific units (pref = -60)
target_IND_Cell{3,nn} = Cell_Inv{5,2,nn}(Cell_Inv{5,3,nn}==3); % specific units (pref = 0)
target_IND_Cell{4,nn} = Cell_Inv{5,2,nn}(Cell_Inv{5,3,nn}==5); % specific units (pref = +60)
end
% Reverse correlation
for target =1:4
% Stimulus parameters
N_image = 2500; % Number of stimulus images
iteration = 100; % Number of iteration
img_size = 227; % Image size
dot_size = 5; % Size of 2D Gaussian filter
% Generate 2D Gaussian filters
[pos_xx,pos_yy] = meshgrid(linspace(1+dot_size,img_size-dot_size,sqrt(N_image)),linspace(1+dot_size,img_size-dot_size,sqrt(N_image)));
pos_xy_list = pos_xx(:) + 1i*pos_yy(:);
[xx_field,yy_field] = meshgrid(1:img_size,1:img_size); xy_field = xx_field + 1i*yy_field;
img_list = zeros(img_size,img_size,3,length(pos_xy_list));
count = 1;
for pp = 1:length(pos_xy_list)
pos_tmp = pos_xy_list(pp);
img_tmp = repmat(exp(-(abs(xy_field-pos_tmp).^2)/2/dot_size.^2)*0.3,1,1,3);
img_list(:,:,:,count) = -img_tmp;
count = count + 1;
end
Gau_stimulus = cat(4,img_list,-img_list);
size(Gau_stimulus)
% Iterative PFI calculation
PFI = zeros(img_size,img_size,3)+255/2; % Initial PFI
PFI_mat = zeros(img_size,img_size,iteration+1); PFI_mat(:,:,1) = PFI(:,:,1);
figure; sgtitle([target_class,'_',num2str(target)]);
set(gcf,'Position', get(0,'Screensize'));
for iter = 1:iteration
tic
disp(num2str(iter));
PFI_0 = PFI; % Save previous PFI
% Generate stimulus as a summation of previous PFI and gaussian stimulus
IMG = repmat(PFI/255,[1,1,1,size(Gau_stimulus,4)])+Gau_stimulus;
IMG = uint8(IMG*255); IMG(IMG<0) = 0; IMG(IMG>255) = 255;
act_collection = [];
for nn=1:NN
net_rand = Cell_Net{nn};
% Measure the response of random AlexNet
act_rand = activations(net_rand,IMG,'relu5'); % Response of 'relu5' layer in random AlexNet
act_reshape = reshape(act_rand,43264,size(IMG,4)); % Reshape the response in 2D form
act_reshape_sel = act_reshape(target_IND_Cell{target,nn},:); % Find the response of face-selective neurons
act_collection = [act_collection; act_reshape_sel];
end
mean_act = mean(act_collection,1); % Average response of face-selective neurons
% Calculate the PFI
norm_act_reshape = repmat(permute(mean_act-min(mean_act),[1,3,4,2]),img_size,img_size,3);
PFI = sum(norm_act_reshape.*double(IMG),4)/sum(mean_act-min(mean_act));
PFI_diff = PFI-PFI_0; PFI = PFI_0 + PFI_diff*10;
PFI(PFI<0) = 0; PFI(PFI>255) = 255;
PFI_mat(:,:,iter+1) = PFI(:,:,1);
subplot(ceil(sqrt(iteration)),ceil(sqrt(iteration)),iter)
imagesc(PFI_mat(:,:,iter+1));axis image off; colormap gray; caxis([0 255]); drawnow
toc
end
PFI_RESULT{target} = PFI_mat(:,:,end);
end
save(filename, 'PFI_RESULT');
else
disp(['Load saved result of PFI analysis ...'])
load(filename);
end
%% Load feature variant stimulus set
[IMG_var, var_cls_idx] = fun_GetStim('invariance_test', var_type, target_class);
IMG_var = single(repmat(permute(IMG_var,[1 2 4 3]),[1 1 3]));
var_idx = var_cls_idx(:, 1); cls_idx = var_cls_idx(:, 2); clearvars var_cls_idx;
figure('units','normalized','outerposition',[0 0.5 1 0.5]); drawnow
sgtitle('Viewpoint-invariant unit and specific units and its visual feature encoding')
%% Visualize PFI and stimulus (Figure 3F)
view_arr = [3, 7, 10];
for ii = 1:4
if ii==4; img_idx = find(cls_idx == 1);
else; img_idx = find(and(var_idx == view_arr(ii), cls_idx == 1));
end
if ii ~= 4
ax1 = subplot(3, 13, ii);
imagesc(IMG_var(:,:,1,img_idx(1)));
colormap(ax1, 'gray');
xticks([]); yticks([]);
end
ax2 = subplot(3, 13, 13+ii);
imagesc(mean(IMG_var(:,:,1,img_idx),4));
colormap(ax2, 'gray');
xticks([]); yticks([]);
ax3 = subplot(3, 13, 26+ii);
imagesc(PFI_RESULT{ii});
colormap(ax3, 'gray');
xticks([]); yticks([]);
switch ii
case 1
ylabel(ax1, 'Example\newline{Stimulus}');
ylabel(ax2, 'Averaged\newline{Stimulus}');
ylabel(ax3, 'PFI');
xlabel(ax3, '-60° spec.');
case 2; xlabel(ax3, '0° spec.');
case 3; xlabel(ax3, '60° spec.');
case 4; xlabel(ax3, 'Invariant\newline{units}');
end
end
%% Correlations between stimulus and PFI (Figure 3G)
stim_correlation = zeros(13, 4);
for kk = 1:4
for ii = 1:13
tmp_corr_mat = zeros(200, 1);
tmp_idx = find(and(var_idx == ii, cls_idx == 1));
for jj=1:200
tmp_image = IMG_var(:,:,1,tmp_idx(jj));
tmp = corrcoef(PFI_RESULT{kk},tmp_image);
tmp_corr_mat(jj) = tmp(1,2);
end
stim_correlation(14-ii,kk) = mean(tmp_corr_mat);
end
end
ax4 = subplot(3, 13, [6 7 8 9 19 20 21 22 32 33 34 35]);
imagesc(stim_correlation);
load('Colorbar_Tsao.mat');
colormap(ax4, cmap);
caxis([-0.02 0.06]);
colorbar;
xline(3.5, '--');
xticks(1:4); xticklabels({'-60° spec.' '0° spec.' '60° spec.' 'Invariant\newline{units}'});
xlabel('Preferred feature image');
yticks(1:2:13); yticklabels({flip(-90:30:90)});
ylabel('Stimulus viewpoint (deg)');
title('Correlations between stimulus and PFI');
%% Combined PFIs of specific units vs. PFIs of invariant units (Figure 3H)
x = 0:0.05:1;
y = 0:0.05:1;
[X,Y] = meshgrid(x,y);
Z=1-X-Y;
C = zeros(size(Z));
for ii = 1:length(x)
for jj = 1:length(y)
xx = x(ii);
yy = y(jj);
zz = 1-xx-yy;
tmp_image = PFI_RESULT{1} * zz + PFI_RESULT{2} * yy + PFI_RESULT{3} * xx;
tmp_image = (tmp_image - min(min(tmp_image)))./(max(max(tmp_image)) - min(min(tmp_image)));
tmp = corrcoef(tmp_image,PFI_RESULT{4});
C(ii,jj) = tmp(1,2);
end
end
ax5 = subplot(3, 13, [11 12 13 24 25 26 37 38 39]);
surf(X, Y, Z, C);
zlim([0 1]);
c = colorbar; colormap(ax5, cmap)
xlabel('a'); ylabel('b'); zlabel('c');
title('Combined PFIs of specific units vs. PFIs of invariant units');