跪求，这段说的怎么写MATLAB代码Then we switch
the source/target pair to get another dataset VOC vs ImageNet.
We uniformly rescale all images to be 256 pixels in length,
and extract 128-dimensional dense SIFT (DSIFT) features
using the VLFeat open source package. A 240-dimensional
codebook is created, where Kmeans clustering is used to
obtain the codewords.我已经有图像矩阵A=2000*1024，