您可以更改trisurf() 返回的 Patch 对象的一些属性:
for i = 1:9
% Compute new values for X, Y, z, and mag
% As an example:
x = linspace(0,1,11-i);
y = x;
[X,Y] = meshgrid(x,y);
mag = hypot(X(:),Y(:));
T = delaunay(X(:),Y(:));
z = i;
Z = z*ones(size(X)); %we could have just called `meshgrid()` with 3 arguments instead
% End recomputation
% Update trisurf() patch: option 1
set( h, 'Faces',T, 'XData',X(T).', 'YData',Y(T).', 'ZData',Z(T).', 'CData',mag(T).' );
pause(0.25); %just so we can see the result
% Update trisurf() patch: option 2
set( h, 'Faces',T, 'Vertices',[X(:) Y(:) Z(:)], 'FaceVertexCData',mag(:) );
pause(0.25); %just so we can see the result
end
假设z 始终是一个标量,就像对trisurf() 的原始调用一样。
- 问:这些选项是否同样快?
- A:我在我的计算机(R2019a,Linux)上运行了一些测试(参见下面的代码),发现当 x/y 位置的数量是 2 到 20 之间的随机数时,多个
set() 调用使用Vertices 可以比使用XData 和相关属性的set() 调用快20%,并且这些策略比多个trisurf() 调用快大约一个数量级。但是,当 x/y 位置的数量允许在 2 到 200 之间变化时,三种方法的运行时间大致相同。
Nruns=1e3;
Nxy_max=20;
for i=1:Nruns
if i==round(Nruns/10)
tic(); %discard first 10% of iterations
end
x = linspace(0,1,randi(Nxy_max-1)+1); %randi([2,Nxy_max]) can be a bit slower
[X,Y,Z] = meshgrid(x,x,randn());
mag = hypot(X(:),Y(:));
T = delaunay(X(:),Y(:));
trisurf(T, X(:), Y(:), Z(:), mag, 'FaceColor', 'interp');
view([-90 90]);
end
tmean_trisurf=1e3*toc()/(Nruns-round(Nruns/10)+1), %in [ms]
h=trisurf(T, X(:), Y(:), Z(:), mag, 'FaceColor', 'interp');
view([-90 90]);
for i=1:Nruns
if i==round(Nruns/10)
tic();
end
x = linspace(0,1,randi(Nxy_max-1)+1);
[X,Y,Z] = meshgrid(x,x,randn());
mag = hypot(X(:),Y(:));
T = delaunay(X(:),Y(:));
set( h, 'Faces',T, 'XData',X(T).', 'YData',Y(T).', 'ZData',Z(T).', 'CData',mag(T).' );
end
tmean_xyzdata=1e3*toc()/(Nruns-round(Nruns/10)+1), %in [ms]
for i=1:Nruns
if i==round(Nruns/10)
tic();
end
x = linspace(0,1,randi(Nxy_max-1)+1);
[X,Y,Z] = meshgrid(x,x,randn());
mag = hypot(X(:),Y(:));
T = delaunay(X(:),Y(:));
set( h, 'Faces',T, 'Vertices',[X(:) Y(:) Z(:)], 'FaceVertexCData',mag(:) );
end
tmean_vertices=1e3*toc()/(Nruns-round(Nruns/10)+1), %in [ms]