Support long video input in seedvr2

[Musisoul]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello, 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 significantly enhances the seedvr2 model's capability to process long video inputs by introducing a robust segmentation and processing pipeline. It also includes performance optimizations across several core tensor operations, ensuring that computations are performed efficiently within the original data types, thereby reducing overhead and improving overall execution speed.

Highlights

• Optimized Tensor Operations: Eliminated unnecessary float() conversions in rms_norm_no_weight, layer_norm_no_weight, and apply_rotary_emb functions, improving performance by keeping computations in the original tensor dtype.
• Long Video Segmentation Support: Introduced a comprehensive pipeline in seedvr_runner.py to segment and process long video inputs, enabling the model to handle videos exceeding previous length limitations.
• Dynamic Video Processing: Implemented new utility functions for probing video metadata, building processing segments with configurable overlap, and reading specific video portions, enhancing flexibility and efficiency for large video datasets.

🧠 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.

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.

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 introduces support for processing long videos in the SeedVR2 super-resolution pipeline by implementing a segmentation-based approach. The video is split into overlapping segments, processed individually, and then stitched back together. The changes also include memory optimizations by avoiding explicit fp32 casting in normalization and RoPE operations, which is crucial for handling large inputs.

My review focuses on the new segmentation logic. While the overlap-save method for stitching segments is a valid approach, the implementation in seedvr_runner.py has some areas for improvement regarding code clarity and design. I've pointed out an unused variable and some possibly unreachable code. More importantly, I've highlighted a pattern of state manipulation that makes the code less modular and harder to maintain. I've also suggested a minor simplification for better readability. Overall, the changes are a significant step towards supporting long videos, and my suggestions aim to improve the robustness and maintainability of the new code.