🧠 Daily Study Log [2025-10-09]
Developed a new idea, continued paper study on InternVideo2, advanced the pose-based dance evaluation research project, and refined resumes for internship applications.

💡 Idea

• Proposed Cup Sense – How Often Should I Wash My Cup?.
• A fun concept visualizing how often a person washes their cup, combining data tracking and gamified feedback to encourage better hygiene habits.

📖 Paper Study – InternVideo2 (CVPR 2024, Day 2)

• Reviewed InternVideo2: Scaling Video Foundation Models for Multimodal Understanding.
• Key Takeaways:
  • Unified spatiotemporal–multimodal architecture combining video, text, and audio.
  • Two main modules:
    • Cross-Modal Fusion Module (CMFM): Enables adaptive information exchange.
    • Hierarchical Temporal Encoder (HTE): Efficiently captures long-term motion.
  • Trains with progressive multitask pretraining — reconstruction, alignment, and generation.
• Findings:
  • Improved temporal reasoning, cross-modal alignment, and video–language generalization.
  • Still requires large-scale data and compute.
• Future Direction:
  • Explore adapter-based fine-tuning and lightweight multimodal extensions.

🧠 Research Progress – Pose-Based Dance Evaluation

• Planned next actions for the dance evaluation study:
  • Collect additional motion data for upcoming experiments.
  • Search potential academic conferences for submission.
  • Research dance generation models for comparative analysis.
• Additional Ideas:
  • Extract optimal BPM and compare with music for alignment accuracy.
  • Measure BPM error by body part to analyze rhythmic precision.
  • Include expert qualitative feedback to identify key skeletal points.
• Implemented a new code module to test these ideas.

📝 Resume Update

• Updated and refined both Korean and English resumes.
• Improved structure, flow, and expression.
• Added recent projects, competition results, and research activities.

✅ TL;DR

📍 Proposed “Cup Sense” – a playful self-tracking hygiene idea
📍 Studied InternVideo2 (Day 2) – unified multimodal video foundation model
📍 Advanced dance evaluation research with new BPM-based analysis
📍 Polished bilingual resume for internship applications