""" An example of using MultiBlockSlice to interleave frames in a virtual dataset. """ import h5py # These files were written round-robin in blocks of 1000 frames from a single source # 1.h5 has 0-999, 4000-4999, ...; 2.h4 has 1000-1999, 5000-5999, ...; etc # The frames from each block are contiguous within each file # e.g. 1.h5 = [..., 998, 999, 4000, 4001, ... ] files = ["1.h5", "2.h5", "3.h5", "4.h5"] dataset_name = "data" dtype = "float" source_shape = (25000, 256, 512) target_shape = (100000, 256, 512) block_size = 1000 v_layout = h5py.VirtualLayout(shape=target_shape, dtype=dtype) for file_idx, file_path in enumerate(files): v_source = h5py.VirtualSource( file_path, name=dataset_name, shape=source_shape, dtype=dtype ) dataset_frames = v_source.shape[0] # A single raw file maps to every len(files)th block of frames in the VDS start = file_idx * block_size # 0, 1000, 2000, 3000 stride = len(files) * block_size # 4000 count = dataset_frames // block_size # 25 block = block_size # 1000 # MultiBlockSlice for frame dimension and full extent for height and width v_layout[h5py.MultiBlockSlice(start, stride, count, block), :, :] = v_source with h5py.File("interleave_vds.h5", "w", libver="latest") as f: f.create_virtual_dataset(dataset_name, v_layout, fillvalue=0)