Heejae Jo , Giyoon Choi, Jonguk Choi, Jongwon Choi. (2025). Improving Generalization Performance of Deep Voice Detection using Semi-Supervised Learning. In Proceedings of the 37th Workshop on Image Processing and Understanding (IPIU 2025 ), Jeju, Korea.

Giyoon Choi, Hyunjin Cho, Heejae Jo , Jonguk Choi, Jongwon Choi. (2024). Knowledge Distillation-Based Prompt Learning Using Attention Decomposition. In Proceedings of the Korean Institute of Electronics Engineers (KIEE) Conference , Jeju, Korea.

Jonguk Choi, Heejae Jo , Taehun Kim, Hyungjun Lim, Jongwon Choi. (2024). Domain Generalization Analysis of Audio Deepfake Detection. In Proceedings of the Korean Institute of Electronics Engineers (KIEE) Conference, Jeju, Korea.

Heejae Jo , et al. (2025). Comparing Face Regions: Reference Image-Guided Explainable Deepfake Detection. Submitted to AAAI 2026 (Reject).

Heejae Jo , et al. (2025). A Partially Manipulated, Explainable Face Synthesis Detection Dataset . In Proceedings of the 2025 Korea Institute of Broadcasting and Media Engineers Autumn Conference, Seoul, Korea. 

Heejae Jo , et al. (2025). Comparing Face Regions: Reference Image-Guided Explainable Deepfake Detection. To be submitted to IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026).

AI Safety: Multi-Modal Anti-Deepfake Framework

• National R&D Program for Trustworthy AI — Mar 2025 ~ Present

• Project Overview
  Designing a comprehensive defect detection system against commonsense reasoning errors. Developing a real-time multi-modal anti-deepfake integrated framework to ensure reliability and safety in AI systems.

• My Contribution
  Created an Audio DeepVoice dataset using Voice Conversion models (Coqui-ai, Seed-VC).
  Built a customized Video-Audio Deepfake dataset optimized for multi-modal detection environments.
  Established a Fake Text Dataset generation pipeline for cross-modal robustness evaluation.
  Developing a Retrieval Deepfake Detector for binary classification and manipulation region segmentation (in progress).

Multi-Projection Space for Large-Scale User Interaction under Low-Light Conditions

• Cultural Technology R&D Program — Mar 2024 ~ Feb 2025

• Project Overview
  Developed technologies for high-degree-of-freedom, large-scale user interaction. Focused on overcoming limitations of low-light illumination environments by leveraging multi-projection spaces.

• My Contribution
  Generated an SMPL dataset using ScoreHMR and structured pose-specific data representations.
  Developed a Crowded Human Detection Model using D-Fine for multi-person motion analysis.

• Oct 2021 - Aug 2023
  Broadcast Operation Team
  CJ OliveNetworks

• Jun 2021 - Oct 2021
  Broadcast Operation Team
  CJ Powercast