feat: migration of natural language image search with a dual encoder to keras 3

[maitry63]

This PR migrates the Natural language image search with a Dual Encoder tutorial from Keras 2 to Keras 3. The implementation is now fully backend-agnostic, allowing it to run seamlessly on JAX, PyTorch, and TensorFlow.

Colab file - Notebook

[gemini-code-assist]

Summary of Changes

Hello @maitry63, 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 modernizes the Natural Language Image Search tutorial by migrating its entire codebase from Keras 2 to Keras 3. The core objective was to achieve full backend-agnosticism, allowing the model to run seamlessly on various deep learning backends like JAX, PyTorch, and TensorFlow. This involved significant updates to data handling, model architecture definitions, and the training loop, ensuring compatibility with the latest Keras API and leveraging keras_nlp for improved text processing.

Highlights

• Keras 3 Migration: The entire tutorial has been migrated from Keras 2 to Keras 3, ensuring compatibility with the latest API and features.
• Backend-Agnostic Implementation: The code is now fully backend-agnostic, allowing it to run seamlessly on JAX, PyTorch, and TensorFlow without modifications.
• Dependency Updates: Removed TensorFlow-specific libraries (tensorflow, tensorflow_hub, tensorflow_text, tensorflow_addons) and integrated Keras 3 core components and keras_nlp for text processing.
• Data Pipeline Refactoring: The data loading and preprocessing pipeline has been refactored from TFRecord-based methods to a custom keras.utils.PyDataset implementation.
• Model Architecture and Training Loop Updates: The project_embeddings utility, create_vision_encoder, create_text_encoder, and DualEncoder classes have been updated to align with Keras 3 API changes, including the compute_loss method for contrastive learning.
• Model Saving Format: Model saving now uses the new .keras extension, which is the recommended format for Keras 3.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

• examples/vision/nl_image_search.py
  • Updated metadata, including author and Keras 3 conversion attribution.
  • Removed TensorFlow-specific imports and added Keras 3 imports (keras, keras.layers, keras.ops, keras.utils, keras_nlp).
  • Adjusted dataset download paths and removed os.remove calls for zip files, streamlining data extraction.
  • Replaced TFRecord-based data processing functions with a new ImageCaptionDataset class, inheriting from keras.utils.PyDataset, for efficient data loading.
  • Modified the project_embeddings utility function to use keras.layers.Activation("gelu") and simplified its parameters.
  • Updated create_vision_encoder to directly use keras.applications.Xception and removed tf.keras.applications.xception.preprocess_input.
  • Refactored create_text_encoder to utilize keras_nlp.models.BertBackbone and accept tokenized inputs (token_ids, padding_mask, segment_ids) instead of raw strings.
  • Rewrote the DualEncoder class's compute_loss method to implement symmetric cross-entropy with normalized embeddings using keras.ops, and removed the explicit train_step and test_step methods.
  • Updated model compilation to use keras.optimizers.AdamW.
  • Removed GPU count logging and adjusted dataset initialization for the new ImageCaptionDataset.
  • Changed model saving to the .keras format.
  • Modified the read_image function to use keras.utils.load_img and img_to_array for image preprocessing.
  • Updated the image embedding generation and find_matches function to align with the new data and model outputs, using np.linalg.norm for normalization.
  • Adjusted the compute_top_k_accuracy function to work with the new data handling and evaluation logic.

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.

Footnotes

1. 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. ↩

[gemini-code-assist]

Code Review

This pull request successfully migrates the Natural Language Image Search example to Keras 3, making it backend-agnostic and updating it to use modern Keras APIs like keras.ops and keras_nlp. A critical security issue was identified regarding the use of insecure HTTP connections for downloading the MS-COCO dataset; updating these URLs to HTTPS is recommended to protect against potential Man-in-the-Middle attacks during data preparation. Additionally, a few suggestions have been provided to address potential issues, such as a missing hyperparameter in the optimizer and a change in the data processing logic that could impact performance.

[maitry63]

@gemini-code-assist review

[gemini-code-assist]

Code Review

This pull request successfully migrates the Natural Language Image Search example to Keras 3, making it backend-agnostic. The changes are extensive and well-executed, including updating the data pipeline from TFRecords to PyDataset, replacing TensorFlow-specific APIs with keras.ops and keras_nlp, and refactoring the DualEncoder model to use the modern Keras 3 compute_loss override. I have a couple of suggestions to improve code clarity and fix a potential bug in the projection head implementation and remove some redundant code in the matching function.

[maitry63]

@gemini-code-assist review

[gemini-code-assist]

Code Review

This pull request successfully migrates the Natural Language Image Search example to Keras 3, leveraging modern features like keras.utils.PyDataset, keras.ops, and keras_nlp for a backend-agnostic implementation. The refactoring also improves the model's architecture and aligns with current best practices. However, a critical security vulnerability has been introduced by globally bypassing SSL certificate verification (ssl._create_unverified_context). This exposes the application to Man-in-the-Middle attacks, especially during data and model downloads. Addressing this security flaw is paramount. Additionally, a couple of minor suggestions are provided for further code enhancement.

[maitry63]

@gemini-code-assist review

[gemini-code-assist]

Code Review

This pull request successfully migrates the Dual Encoder image search tutorial to Keras 3, making it backend-agnostic and leveraging KerasNLP for the text encoder. However, a critical security issue was identified where SSL certificate verification is explicitly disabled during data download, which could expose users to Man-in-the-Middle attacks. Additionally, there's a significant change in the architecture of the projection head that likely deviates from the original intent, and some performance redundancies exist in the loss computation and inference steps.

[maitry63]

@gemini-code-assist review

[gemini-code-assist]

Code Review

This pull request successfully migrates the Natural Language Image Search example to Keras 3, making it backend-agnostic and updating it to use modern Keras APIs like keras.utils.PyDataset and keras_nlp. However, a critical security vulnerability was identified in the data download logic where SSL certificate verification is explicitly disabled, exposing it to Man-in-the-Middle attacks. This issue must be addressed to ensure secure data downloads.

[sachinprasadhs]

I have added my comments, please address, for the resolved Gemini comments, please mark them as resolved.