Listening to the Unspoken: Exploring "365" Aspects of Multimodal Interview Performance Assessment

[MM 2025] The official implementation for the paper titled: "Listening to the Unspoken: Exploring '365' Aspects of Multimodal Interview Performance Assessment"

🏆 Championship Solution of ACM Multimedia AVI Challenge 2025 Track 2: Interview Performance Assessment

🎯 Project Overview

The task of Interview Performance Assessment is a multi-input, multi-label regression task. Given videos in which subjects respond to both generic and personality questions, the objective is to develop models and algorithms to evaluate five job-related competencies:

• Integrity
• Collegiality
• Social versatility
• Development orientation
• Overall hireability

📊 Leaderboard

Team Name	MSE (↓)
HFUT-VisionXL (our)	0.18240 (1)
CAS-MAIS	0.18510 (2)
ABC-Lab	0.19394 (3)
The innovators	0.20749 (4)
HSEmotion	0.22150 (5)
USTC-IAT-United	0.24828 (6)
DERS	0.25540 (7)

🧱 Model Framework

Figure 1: The overall model framework.

Figure 2: Multimodal Shared Compression Multilayer Perceptron (MSCMLP)

⚙️ Directory Structure

track2/
├─ args_log/                           # Stores parameter configs for each experiment run
│  └─ .gitkeep
├─ data/
│  ├─ all_data.csv                     # CSV file containing both training and test data
│  ├─ test_data_basic_information.csv  # Test set CSV file
│  ├─ train_data.csv                   # Training set CSV file
│  ├─ val_data.csv                     # Validation set CSV file
│  └─ val_data_new.csv                 # CSV file with redundant columns removed
├─ dataset/
│  └─ baseline_dataset2_vote.py        # Dataset class for track2
├─ img/                                # Stores loss curve plots
│  └─ .gitkeep
├─ model/
│  └─ vote_model/
│     └─ M_model.py                    # Model implementation
├─ train_print_log/
│  └─ .gitkeep
├─ .gitignore
├─ README.md
├─ requirement.txt
├─ train_task2_vote.py
└─ vote_train.sh                       # Script to run training

📋 License

This project is licensed under the MIT License - see the LICENSE file for details.

📞 Contact

If you have any questions or suggestions, please contact the project maintainers (HFUT-VisionXL).

---

⚠️ Note: This project is for academic research purposes only. Please comply with relevant data usage agreements and competition rules.

🙏 Acknowledgments

• 🏆 Thanks to the AVI Challenge 2025 organizers
• 🤗 Thanks to the developers of MERtools for their excellent open-source tools that supported our data preprocessing.