Multiple crops example

[mzient]

Category:

Other (e.g. Documentation, Tests, Configuration)

Description:

Implements a pipeline example requested by a user in #1579

Additional information:

Affected modules and functionalities:

N/A

Key points relevant for the review:

Tests:

• Existing tests apply
• New tests added
  • Python tests
  • GTests
  • Benchmark
  • Other
• N/A

Checklist

Documentation

• Existing documentation applies
• Documentation updated
  • Docstring
  • Doxygen
  • RST
  • Jupyter
  • Other
• N/A

DALI team only

Requirements

• Implements new requirements
• Affects existing requirements
• N/A

REQ IDs: N/A

JIRA TASK: N/A

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[dali-automaton]

CI MESSAGE: [18958041]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [18958465]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [18958465]: BUILD PASSED

[stiepan]

For the future, I belive we could make the resize and crop ops support similar trick, likely in a bit more straightforward way.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

szkarpinski commented on 2024-10-04T11:44:12Z
----------------------------------------------------------------

Line #42.        )  # convert to float and mark outermost dimension as"frame"

Nitpicks: Why +0.0 and not cast? Also, there's a missing space before "frame"

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

szkarpinski commented on 2024-10-04T11:44:13Z
----------------------------------------------------------------

How about adding a sentence explaining what the images above mean? The first single image was confusing to me - I expected sequences of equal length. It took me some time to notice num_crops = random (...) in the code.

mzient commented on 2024-11-04T11:46:51Z
----------------------------------------------------------------

Yeah, I'd recommend playing with the seed a bit to get an example with three different sequence lengths, preferably without a single-image sequence.

[mzient]

Yeah, I'd recommend playing with the seed a bit to get an example with three different sequence lengths, preferably without a single-image sequence.

---

View entire conversation on ReviewNB

[greptile-apps]

Greptile Overview

Greptile Summary

Adds a new Jupyter notebook example demonstrating how to extract multiple crops from a single image using DALI's fn.warp_affine, fn.per_frame, and fn.transforms.crop operators. This addresses a user request from issue #1579.

• Two pipeline examples are provided: equal-sized crops and variable-sized crops resized to a common output size
• Both techniques leverage the per-frame broadcasting behavior of fn.warp_affine when its arguments are marked with fn.per_frame
• Documentation index updated with appropriate operator references
• Minor style issues: typo in intro text and unused imports (math, matplotlib.patches)

Confidence Score: 5/5

• This PR is safe to merge - it only adds documentation/example code with no impact on production functionality.
• Documentation-only change adding a Jupyter notebook example. The code follows established patterns in the DALI examples and uses existing DALI operators correctly. Only minor style issues found.
• No files require special attention. The notebook example could have unused imports removed but this is non-blocking.

Important Files Changed

File Analysis

Filename	Score	Overview
docs/examples/image_processing/index.py	5/5	Added new doc_entry for multiple_crops.ipynb with operator references for fn.warp_affine, fn.transforms.crop, and fn.per_frame. Clean addition following existing patterns.
docs/examples/image_processing/multiple_crops.ipynb	4/5	New Jupyter notebook demonstrating multiple crops from a single image. Has unused imports (math, patches) and a typo "diplay". Core pipeline logic is correct.

Sequence Diagram

sequenceDiagram
    participant User as User/Pipeline
    participant FileReader as fn.readers.file
    participant Decoder as fn.decoders.image
    participant Shapes as fn.shapes
    participant Random as fn.random.uniform
    participant Stack as fn.stack
    participant PerFrame as fn.per_frame
    participant Transforms as fn.transforms.crop
    participant WarpAffine as fn.warp_affine

    User->>FileReader: Read image files
    FileReader->>Decoder: JPEG bytes
    Decoder->>Shapes: Get image shape (H, W)
    Shapes->>Random: Generate crop params using dimensions
    Random->>Stack: Stack crop coordinates
    Stack->>PerFrame: Mark as per-frame data
    PerFrame->>Transforms: Create crop transform matrix
    Decoder->>WarpAffine: Input image
    Transforms->>WarpAffine: Per-frame transform matrices
    WarpAffine->>User: Output sequence of crops

Loading

[greptile-apps]

Additional Comments (2)

1. docs/examples/image_processing/multiple_crops.ipynb, line 1 (link)

   style: Typo: "diplay" should be "display"

2. docs/examples/image_processing/multiple_crops.ipynb, line 26-27 (link)

   style: Unused imports: matplotlib.patches as patches and math are imported but never used in the notebook.

_{2 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}