New function [CSVSink] - allowing to serialise Detections to a CSV file

[AdonaiVera]

Description

This PR introduces the CSVSink class to address structured logging of object detection data into CSV format, as requested in issue #746. The class supports logging of bounding boxes, class IDs, confidence scores, and frame numbers, enhancing post-processing data analysis and interpretability.

The PR includes:

• CSVSink class implementation.
• Comprehensive unit tests for the class.
• A demo showcasing the integration of CSVSink with a video processing pipeline.

Note: The CSVSink class was placed directly in the utils folder in the file.py file due to its capability to handle various types of file operations integral to the project's core functionality. Open to suggestions if there's a more appropriate location within the project structure.

Type of Change

• New feature: Added CSVSink class for structured logging of detection data.

Testing

The CSVSink class was validated through a series of unit tests, covering:

• Logging with various custom data scenarios.
• Handling of scenarios with no detections.
• Verification of CSV output format and data integrity.

A demo script demonstrates the class's usage within a video processing workflow, annotating video frames and logging detections to a CSV file. Demo Here

Documentation Updates

• Added CSVSink class documentation within the module, detailing its purpose, usage, and example usage.
• Updated README to include a section on structured detection data logging, introducing the CSVSink class.

[SkalskiP]

Hi @AdonaiVera 👋🏻 ! Good start. I left a few comments. Let me know if you have any questions. I didn't look at unit tests yet. Let's work on logic first.

[AdonaiVera]

Hi @SkalskiP

Thank you for your valuable feedback. I have already implemented the changes and created a PR. Additionally, I updated the demo and moved the unit test to the test/detections folder. Please let me know your thoughts on the changes and if you have any further suggestions 🥷

[SkalskiP]

As I already mentioned. For the method/function to be visible in docs. It needs to be public, and it needs to have docstring. CSVSink methods are undocumented, and as a result, users don't see them in docs.

I recommend using mkdocs serve locally to preview your docs changes.

[AdonaiVera]

Hi @SkalskiP 👋
I have made all the necessary changes. However, there are still some questions that I left on the thread.

I tried to handle multiple scenarios, but I can work with more if needed. But before that, I need to solve the question about what happens if the custom data changes between detections.

Thanks 🚀

[SkalskiP]

I made a few docs updates in the meantime. Nothing important. There was no point in forcing you to make corrections. These were very minor things.

[SkalskiP]

Hi @AdonaiVera 👋🏻 I made two small commits to your branch today. I really wanted to close it out before the weekend. Feel free to ask questions here in comments.

Overal awesome work! Thanks a lot for your help! 🔥

[AdonaiVera]

Thank you! We did an amazing collaborative work. @SkalskiP 💪

Now that I have the full perspective, I'm ready to create the JSON saving function, I'll update that PR soon 🚀

[SkalskiP]

Awesome @AdonaiVera! Cant wait to see that! 🔥