Discussion on Video Processing Architecture & Contribution Areas Across OpenVINO Repositories

[AdityaSinghDevs]

Hi OpenVINO team,

I've been contributing to Anomalib and Datumaro, recently working on:

• Metric logging improvements (PR #2489 in Anomalib)
• Implementing state management and core set frame list dumping for video processing and near-duplicate removal
  functionality (PR #1697 in Datumaro)

Through these contributions, I’ve noticed some interesting patterns in how video frame processing is handled across different repositories and wanted to open a discussion around this.

A few technical questions have emerged:

1. Architecture & Standardization

• Is there a preferred architectural approach for large-scale video processing across OpenVINO’s ecosystem?
• Are there ongoing efforts to standardize video processing methodologies between Anomalib, Datumaro, and other related projects?

2. Cross-Repository Integration

• There seem to be overlapping functionalities in video frame handling across repositories. Is there a vision for more integration or shared utilities?
• How do smaller repositories like Anomalib and Datumaro fit into the broader OpenVINO ecosystem in the long term?

3. Future Contribution Areas

• Are there specific gaps or areas within video processing that could use more contributions?
• Would it be helpful to document some of the key learnings and patterns I’ve come across while implementing these features?

I’d love to get insights from the team and the community to better align my contributions with the overall direction of the project. Looking forward to your thoughts!

Thanks,
Aditya