Fix: handle 3D input for channels_first in preprocess_input() #21749 #21754

utsab345 · 2025-10-17T12:15:13Z

Summary

Fixed a shape mismatch issue when preprocess_input() receives 3D input (single image)
with data_format='channels_first'. The bug occurred because the mean and std tensors
were reshaped for 4D input only.

Changes

Added conditional reshape for both 3D and 4D cases.
Included shape validation for unsupported dimensions.

Test

import numpy as np, tensorflow as tf, keras
image = tf.transpose(tf.random.uniform((224, 224, 3)), perm=[2, 0, 1])
keras.applications.vgg16.preprocess_input(image, data_format='channels_first')

Closes #21749

…andling and tests

…21749)

gemini-code-assist · 2025-10-17T12:15:31Z

Summary of Changes

Hello @utsab345, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request primarily addresses a critical bug in the preprocess_input function within Keras applications, ensuring it correctly handles 3D image inputs when using a channels_first data format. Beyond the core fix, it also enhances the robustness of image utility functions by improving file format handling and adding new tests for image saving.

Highlights

Bug Fix: preprocess_input for 3D channels_first: Fixed a shape mismatch issue in keras.applications.vgg16.preprocess_input() when receiving 3D input (single image) with data_format='channels_first', which previously occurred because mean and std tensors were only reshaped for 4D input.
Conditional Reshaping and Validation: Implemented conditional reshaping logic for mean and standard deviation tensors within _preprocess_tensor_input to correctly handle both 3D and 4D inputs when data_format is channels_first. Additionally, added explicit shape validation to raise a ValueError for unsupported input dimensions.
Improved save_img Utility: Enhanced the save_img utility by normalizing file_format='jpg' to file_format='jpeg' and updating the warning message for RGBA to RGB conversion to reflect the 'JPEG' format explicitly.
New Integration Tests for save_img: Added new integration tests for the save_img utility to ensure proper saving of JPG images, including verification of RGBA to RGB conversion when saving to the JPEG format.
.gitignore Update: The .gitignore file was updated to include the venv directory, preventing virtual environment files from being committed.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

codecov-commenter · 2025-10-17T12:17:21Z

Codecov Report

❌ Patch coverage is 21.42857% with 11 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.83%. Comparing base (e2be4de) to head (a935e1c).

Files with missing lines	Patch %	Lines
keras/src/applications/imagenet_utils.py	27.27%	4 Missing and 4 partials ⚠️
keras/src/utils/image_utils.py	0.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #21754      +/-   ##
==========================================
+ Coverage   82.63%   84.83%   +2.20%     
==========================================
  Files         572      572              
  Lines       58555    58566      +11     
  Branches     9153     9158       +5     
==========================================
+ Hits        48385    49687    +1302     
+ Misses       7843     6433    -1410     
- Partials     2327     2446     +119

Flag	Coverage Δ
keras	`84.65% <21.42%> (+2.21%)`	⬆️
keras-jax	`65.43% <21.42%> (+2.23%)`	⬆️
keras-numpy	`57.66% <0.00%> (+0.09%)`	⬆️
keras-openvino	`34.49% <0.00%> (+0.11%)`	⬆️
keras-tensorflow	`66.19% <21.42%> (+2.23%)`	⬆️
keras-torch	`65.72% <21.42%> (+2.21%)`	⬆️
keras.applications	`83.31% <27.27%> (?)`
keras.applications-jax	`83.31% <27.27%> (?)`
keras.applications-numpy	`22.72% <0.00%> (?)`
keras.applications-openvino	`22.72% <0.00%> (?)`
keras.applications-tensorflow	`83.31% <27.27%> (?)`
keras.applications-torch	`83.04% <27.27%> (?)`

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

gemini-code-assist

Code Review

This pull request correctly fixes a shape mismatch issue in preprocess_input for 3D inputs with data_format='channels_first'. The changes are well-tested and also include a nice improvement in save_img to normalize the 'jpg' format. I've suggested a couple of improvements in imagenet_utils.py to improve the error messages and reduce code duplication. Overall, this is a good contribution.

gemini-code-assist · 2025-10-17T12:18:15Z

keras/src/applications/imagenet_utils.py

+        if ndim == 3:
+            mean_tensor = ops.reshape(mean_tensor, (3, 1, 1))
+        elif ndim == 4:
+            mean_tensor = ops.reshape(mean_tensor, (1, 3, 1, 1))
+        else:
+            raise ValueError(f"Unsupported shape for channels_first: {x.shape}")


While this logic correctly handles 3D and 4D inputs, the error message could be more informative. According to the Keras API design guidelines, a good error message should explain what was expected and how the user can fix it.¹ This message only states what was received. Consider clarifying that only 3D and 4D tensors are supported for channels_first.

Additionally, you can make the code slightly cleaner by checking for the invalid ndim case first.

Suggested change

if ndim == 3:

mean_tensor = ops.reshape(mean_tensor, (3, 1, 1))

elif ndim == 4:

mean_tensor = ops.reshape(mean_tensor, (1, 3, 1, 1))

else:

raise ValueError(f"Unsupported shape for channels_first: {x.shape}")

if ndim not in (3, 4):

raise ValueError(

f"Unsupported tensor rank: {ndim}. With `data_format='channels_first'`, "

"`preprocess_input` only supports 3D (single image) and 4D (batch of "

f"images) tensors. Received tensor with shape: {x.shape}"

)

if ndim == 3:

mean_tensor = ops.reshape(mean_tensor, (3, 1, 1))

elif ndim == 4:

mean_tensor = ops.reshape(mean_tensor, (1, 3, 1, 1))

Style Guide References

Footnotes

The style guide states that error messages should be contextual, informative, and actionable, explaining what happened, what was expected, and how to fix it. ↩

gemini-code-assist · 2025-10-17T12:18:15Z

keras/src/applications/imagenet_utils.py

        if data_format == "channels_first":
-            std_tensor = ops.reshape(std_tensor, (-1, 1, 1))
+            if ndim == 3:
+                std_tensor = ops.reshape(std_tensor, (3, 1, 1))
+            elif ndim == 4:
+                std_tensor = ops.reshape(std_tensor, (1, 3, 1, 1))
+            else:
+                raise ValueError(
+                    f"Unsupported shape for channels_first: {x.shape}"
+                )
+        else:
+            std_tensor = ops.reshape(std_tensor, (1,) * (ndim - 1) + (3,))


This logic for reshaping std_tensor is a duplicate of the logic used for mean_tensor above. To improve maintainability and reduce code duplication, consider refactoring this. You could determine the reshape_shape once at the start of the data_format == 'channels_first' block and reuse it for both tensors. This would also apply to the error handling logic.

fchollet · 2025-10-17T23:24:48Z

keras/src/utils/image_utils.py

    """
    data_format = backend.standardize_data_format(data_format)
+    # Normalize jpg → jpeg
+    if file_format is not None and file_format.lower() == "jpg":


Please revert changes in this file

fchollet · 2025-10-17T23:25:20Z

integration_tests/test_save_img.py

+        ((50, 50, 4), "rgba.jpg"),
+    ],
+)
+def test_save_jpg(tmp_path, shape, name):


Please use a TestCase subclass and use self.assert... methods instead of naked assert staments

fchollet

Thanks for the PR!

utsab345 and others added 7 commits October 4, 2025 18:00

Simplify save_img: remove _format, normalize jpg→jpeg, add RGBA→RGB h…

2b65f89

…andling and tests

Merge branch 'keras-team:master' into master

2a5bb21

Merge branch 'keras-team:master' into master

6e0340b

Merge branch 'keras-team:master' into master

7b84f95

Fix bug in preprocess_input for 3D channels_first inputs (keras-team#…

49fc1b7

…21749)

Fix: handle 3D input for channels_first in preprocess_input()

544e7f1

Fix: handle 3D input for channels_first in preprocess_input()

a935e1c

google-ml-butler bot added the size:M label Oct 17, 2025

google-ml-butler bot assigned gbaned Oct 17, 2025

gemini-code-assist bot reviewed Oct 17, 2025

View reviewed changes

fchollet reviewed Oct 17, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix: handle 3D input for channels_first in preprocess_input() #21749 #21754

Fix: handle 3D input for channels_first in preprocess_input() #21749 #21754

Uh oh!

utsab345 commented Oct 17, 2025

Uh oh!

gemini-code-assist bot commented Oct 17, 2025

Uh oh!

codecov-commenter commented Oct 17, 2025 •

edited

Loading

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Oct 17, 2025

Uh oh!

gemini-code-assist bot Oct 17, 2025

Uh oh!

fchollet Oct 17, 2025

Uh oh!

fchollet Oct 17, 2025

Uh oh!

fchollet left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Fix: handle 3D input for channels_first in preprocess_input() #21749 #21754

Are you sure you want to change the base?

Fix: handle 3D input for channels_first in preprocess_input() #21749 #21754

Uh oh!

Conversation

utsab345 commented Oct 17, 2025

Summary

Changes

Test

Closes #21749

Uh oh!

gemini-code-assist bot commented Oct 17, 2025

Summary of Changes

Highlights

Footnotes

Uh oh!

codecov-commenter commented Oct 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Oct 17, 2025

Choose a reason for hiding this comment

Style Guide References

Footnotes

Uh oh!

gemini-code-assist bot Oct 17, 2025

Choose a reason for hiding this comment

Uh oh!

fchollet Oct 17, 2025

Choose a reason for hiding this comment

Uh oh!

fchollet Oct 17, 2025

Choose a reason for hiding this comment

Uh oh!

fchollet left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

codecov-commenter commented Oct 17, 2025 •

edited

Loading