matlab 读取图片的格式

在matlab中读取图像数据用imread函数：img = imread(‘******png’)

其中img的格式为：height x width x channels x num ; 3通道的顺序为RGB

所以如果需要在matlab中对图像数据做一些转换，转换成caffe或者C所支持的格式，可以用如下代码：

% Convert to BGR
images = images(:,:,[3 2 1],:);
% Switch width and height
images = permute(images, [2 1 3 4]);

附上deep-metric里面的一段代码：

%% Pick image pairs
assert(mod(batch_size, 2)==0);
[image_id_pairs, labels] = get_training_examples_multilabel(mode, batch_size); % get_training_examples_multilabel只是针对class id和image id进行一系列的操作，并未涉及到真正的图像数据

%% Prep dataset for C and Caffe conventions
images = cat(4, images.img); %%Concatenate arrays along specified dimension

assert(max(image_id_pairs(:)) <= length(images));
% images must be (height x width x channels x num)
assert(ndims(images) == 4);
% assume square images
assert(size(images, 1) == size(images, 2));
% images must have 3 channels
assert(size(images, 3) == 3);

% Convert to BGR
images = images(:,:,[3 2 1],:);
% Switch width and height
images = permute(images, [2 1 3 4]);

% Caffe likes 0-1 labels
labels = int32(labels);
labels(labels == -1) = 0;

labels_cont = 1:11318;
assert(isequal(double(unique(labels)), labels_cont'));

% Subtract 1 for C indexing
image_id_pairs = int32(image_id_pairs - 1);

% check the number of paired data is divisible 

另外，python中 cv2.imread读取的图像的存储格式为H W K，且三通道的顺序为BGR ;

PIL.Image.open读取图片的存储格式为：W H K ，且三通道顺序为RGB