Matlab写入H5文件后Python(h5py)读取轴顺序变化问题

This boils down to a core difference in array storage conventions between MATLAB and Python (including h5py), and how each tool interacts with the HDF5 standard:

1. MATLAB uses column-major order, h5py uses row-major #

MATLAB is built around column-major (Fortran-style) array storage, while Python (and most C-based tools like h5py) uses row-major (C-style). HDF5 itself follows row-major order, but MATLAB automatically adjusts for this mismatch during write/read operations:

• When you save your 64×64×3×4 array to HDF5 in MATLAB, it internally transposes the dimensions to 4×3×64×64 to align with HDF5's storage rules.
• When you read the file back in MATLAB, it quietly transposes the dimensions again to restore your original 64×64×3×4 shape—this happens behind the scenes, so you never see the transposed version.

2. Confirm with HDF5 metadata #

If you run h5disp('data.h5') in MATLAB, you’ll see the actual stored dimensions of /Image are 4×3×64×64. h5py reads the file exactly as it’s stored, so it returns that shape directly without any automatic adjustment.

3. Fix the shape in Python #

To get the same dimension order as MATLAB, just transpose the array after reading it with h5py. For your specific case, you can explicitly map the axes to reverse the transposition:

import h5py

hf = h5py.File('./data.h5', 'r')
syn1 = hf['/Image'][()]  # Read the dataset into a NumPy array
# Convert 4×3×64×64 back to 64×64×3×4
syn1_corrected = syn1.transpose((2, 3, 1, 0))

This will give you an array with the exact dimension order you had in MATLAB.

内容的提问来源于stack exchange，提问作者bhushan23