This week I focused on preparing the LumiKey system for the upcoming demo by improving the stability and reliability of the full software pipeline. Building on last week’s work validating the end-to-end flow, the goal this week was to make sure the system behaves consistently under demo conditions and handles errors more gracefully.
Pipeline Stability Improvements
A major focus this week was improving the consistency of the full pipeline: image/PDF input → preprocessing → OMR → MusicXML cleaning → conversion → note events. While the pipeline was already functional, I worked on making sure it produces stable outputs across repeated runs and different input types.
I refined preprocessing parameters to better handle difficult cases such as skewed images, uneven lighting, and PDFs. I also made adjustments to ensure that intermediate outputs passed between stages are consistently formatted, which helped reduce unexpected parsing or conversion issues. These changes improved the overall reliability of the pipeline when tested on the benchmark dataset.
Debugging and Error Handling
To better prepare for demo scenarios, I improved error handling throughout the pipeline. The system now clearly reports which stage fails (preprocessing, OMR, MusicXML cleaning, or parsing) instead of failing silently. This makes it easier to diagnose issues quickly during live demonstrations.
I also addressed several edge cases discovered during testing, including:
- OMR producing no output file
- Invalid or empty MusicXML files
- Inconsistent outputs from certain image inputs
By handling these cases explicitly, the system is now more robust and less likely to crash or produce confusing results during the demo.
End-to-End Testing for Demo Inputs
I continued testing the pipeline using the benchmark dataset and focused on selecting a subset of reliable demo inputs. These inputs were chosen based on consistent performance across preprocessing, OMR, and conversion stages.
I also verified that the outputs from the pipeline align with the expected structure of note events, ensuring they can be used directly by the playback system and hardware components.
Integration Readiness
In preparation for demo integration, I ensured that the event output format remains consistent and compatible with the existing playback and hardware systems. This helps ensure smooth communication between the software pipeline and the ESP32 during the demo.
I also worked with the team to verify that the full flow—from sheet music input to event generation—can be demonstrated clearly and reliably.
Schedule
The project remains on schedule and ready for the upcoming demo. The software pipeline is now stable enough to support live demonstrations, and remaining work is focused on final integration and polishing.
Deliverables for Next Week (Demo Execution and Final Refinements)
Next week I will focus on supporting the demo on Monday and Wednesday and addressing any issues that arise during live testing. After the demo, I plan to continue refining the pipeline based on observed edge cases, especially improving robustness for more diverse inputs.
I will also continue working with the team on deeper integration with the hardware system and identifying any improvements needed for timing, accuracy, or user experience as we move toward the final system.