🧠 Daily Study Log [2025-07-25]
A multi-threaded day of ideation, modeling, implementation, and reading.
Progressing steadily toward motion feedback systems and competition optimization.

💡 Project Idea — Sweat Forecast: Predicting Pit Stain Risk by Hour

Proposed a lighthearted but data-driven idea to predict underarm sweat visibility based on weather, clothing, and user activity.

Key Points:

• Binary classifier for “Is it safe to wear gray?”
• Use weather, clothing color, activity level as features
• Visualize hourly risk via heatmaps
• Future app/web interface for “pit stain risk” alerts
• Strong use case in summer fashion decisions

📝 TOEIC Practice

Studied:

• Listening Part 3
• Reading comprehension

Keeping consistent progress in test prep.

🍓 Team — Smart Farm Strawberry Competition

Participated in team meeting to discuss concept refinement.
Reviewed key components for smart farming applications.

🕺 Personal Project — PoseSyncEvaluator: 2D Pose-Based Dance Feedback

Wrapped up initial implementation for pose-based motion comparison system.
Successfully tested locally in VS Code after running feature extraction on Colab.

Highlights:

• Used MediaPipe for 2D keypoint extraction
• DTW + cosine similarity for alignment and scoring
• Streamlit interface launched on local machine
• All frame .npy data downloaded from Colab and configured properly

This marks the first real deployment of a full feedback pipeline using pose data.

📄 Paper Review — Explainable Quality Assessment for Martial Arts

✅ Day 2 – Literature & Method Review

Focused on methods for aligned motion scoring:

• Covered pose input types and evaluation paradigms
• Explored Procrustes + DTW alignment
• Reviewed ensemble regression techniques
• SHAP used for joint-level interpretability
• Noted that alignment quality is more crucial than model complexity

Planned next step: implement 2D version of skeleton angle extraction + scoring.

🏆 Competition — Electricity Usage Forecasting (DACON)

Ran 3 submissions today with various tuning strategies:

Next Plan:

• Try SelectFromModel + LightGBM
• Explore Stacking Ensemble w/ top features
• Segment models by building type
• Add lag-based temporal features

✅ TL;DR

📍 Idea: Sweat Forecast — pit stain risk predictor using weather/activity
📍 TOEIC: Part 3 listening + reading
📍 Team: Strawberry smart farm discussion
📍 Project: PoseSyncEvaluator deployed locally (Streamlit success)
📍 Paper: Studied SHAP + DTW-based scoring method
📍 Competition: 3 new submissions (best SMAPE: 11.657)
📍 Next: Scoring model implementation + feature enhancement