This week I focused on finalizing and stabilizing the Scan Layer and overall software pipeline in preparation for the demo. As I began testing more complex sheet music inputs, especially larger files and higher-resolution PDFs, I encountered issues where the system would produce unreliable or blank outputs. My main goal this week was to resolve these issues and ensure consistent behavior across a wider range of inputs.
Pipeline Refinement for Complex Inputs
While testing more complex sheet music, I found that certain inputs caused failures in the OMR stage. In particular, large or high-resolution files sometimes exceeded what the OMR engine could handle, and lower-quality inputs occasionally failed to produce usable results.
To address this, I restructured parts of the Scan Layer pipeline to better balance input quality and processing reliability. I introduced normalization steps for both image and PDF inputs before they are passed into the OMR stage, ensuring that inputs fall within a more consistent and OMR-friendly range. This included resizing inputs to appropriate resolutions and standardizing preprocessing behavior across different file types.
I also improved how the system handles cases where the OMR output is incorrect but not completely missing. In some cases, notes were misinterpreted in unrealistic pitch ranges, which caused incorrect playback. To mitigate this, I added a correction step that adjusts clearly invalid note ranges back into a realistic piano range. This ensures that even imperfect OMR outputs remain usable for downstream playback.
These changes significantly improved the consistency of both image and PDF processing and reduced cases where the pipeline produced no output.
Improved Robustness and User Feedback
In addition to backend improvements, I added better user feedback during the scan process. The system now previews uploaded images and PDFs before processing, allowing users to confirm that the correct file was selected. This helps prevent user error and improves overall usability.
I also ensured that failures are more clearly handled, so that if the pipeline encounters an issue, it does not silently fail but instead provides more visible feedback.
Verification and Stability
After these updates, I reran the verification suite to ensure the pipeline remained stable. The tests passed successfully, confirming that the changes did not introduce regressions and that the system continues to meet the required correctness and performance criteria.
Testing across a wider range of inputs also gave more confidence that the system behaves reliably under more realistic conditions, not just ideal cases.
Demo Preparation
For demo preparation, I tested the full pipeline multiple times using a variety of inputs to ensure consistent behavior. I identified a set of reliable demo cases and verified that they produce correct outputs and smooth playback.
I also reviewed the full system flow and ensured I am prepared to explain how the Scan Layer integrates with the rest of the pipeline, as well as the key design decisions and trade-offs.
Schedule
The project is on schedule and ready for the demo. The system is now more robust, handles a wider range of inputs, and has been verified against key requirements.
Deliverables for Demo
For the demo, the system will show the full pipeline from sheet music input to LED guidance and playback. I will focus on demonstrating system reliability, correctness of note generation, and responsiveness, while also being prepared to discuss the challenges encountered and how they were resolved.