Skip to content

Dataset Chunking

Overview

The Dataset Chunking tool in MIB allows you to split a large dataset into smaller pieces and later restore them. This feature is useful for parallel segmentation of large datasets across multiple workstations.

Scheme of the chopped images workflow

Dataset Chunking -> Chunk dataset

Chopped images -> Export Dialog

The Chunk dataset command chops a large dataset into smaller pieces.

Settings

  • Number of tiles in X/Y/Z: define the number of resulting datasets
Example

Setting X: 2, Y: 1, Z: 1 splits the dataset into two parts: [1:width/2] and [width/2:width], where width is the original dataset’s width.

  • : when checked, chops the model layer into corresponding blocks.
  • : when checked, chops the Mask layer into corresponding blocks.
  • ...Output directory: specify the directory for saving chopped datasets (via button or text input).
  • Filename template: sets the naming pattern; MIB appends _Znn_Xnn_Ynn (e.g., _Z01_X01_Y01) to each block’s filename, where nn is the block index.

  • Output format for images: choose from

    • Amira Mesh
    • NRRD
    • 3D-TIF
    • HDF5 with XML header

  • Output format for models: options include MATLAB, Amira Mesh, NRRD, TIF, or HDF5; saved with a Labels_ prefix.

Masks are saved in MATLAB format with the template Mask_[FN].mask, where [FN] matches the corresponding image filename.

Dataset Chunking -> Stitch dataset

Chopped images -> Import Dialog

The Stitch dataset command restores a previously chopped dataset, while Fuse into dataset fuses cropped datasets into the currently open dataset. Both open the same dialog, preset to the Generate new stack or Fuse into existing mode respectively.

Modes

Mode Description
Generate new stack Combines images, models, or masks into a new stack. Requires filenames with _Znn_Xnn_Ynn tags for images and Labels_ prefixes for models (default for chopped exports).
Fuse into existing Merges selected datasets into the current one using BoundingBox data from the ImageDescription field. Useful for cropped datasets; includes X/Y/Z offset fields (in pixels). On a BigData dataset, a model saved at a lower resolution is fused into the matching pyramid level; the offsets remain in full-resolution pixels.

Steps

  1. Choose a mode and select the types to combine (, , ).
  2. Press Select files to pick files.

Filename Generation

  • Image: Huh7_CmVTag1_R2_Pos5_crop_chop_Z01-X01-Y01.am
  • Model: Labels_Huh7_CmVTag1_R2_Pos5_crop_chop_Z01-X01-Y01.model
  • Mask: Mask_Huh7_CmVTag1_R2_Pos5_crop_chop_Z01-X01-Y01.mask

Important!

If is checked, select only image files (not models or masks); filenames for models/masks are auto-generated. For importing models/masks into an open dataset (without ), select the actual model/mask files.

Usage Tips

Parallel Segmentation

To segment a large dataset across workstations:

  1. Export with desired tiles (e.g., X: 2, Y: 2, Z: 1).
  2. Process each chunk independently.
  3. Import with Generate new stack to recombine.
Offset Adjustment

Use Fuse into existing with X/Y/Z offsets to align cropped datasets that don’t match the default BoundingBox.

Back to MIB | User interface | Ribbon | Home