SCRUM-43: Integrated MesoNet into Ensemble

[frank-bi-canada]

Jira Ticket

SCRUM-43: Integrated MesoNet into Ensemble

Summary

MesoNet can now take faces from video_handler.py and output a result. Currently the result is unused.

Due to the large differences in the virtual environments of the ensemble and MesoNet, communication between the two is facilitated through HTTP requests. The Python 3.6 mesonet environment will be run on a server (current default of port 8000) and provide a client that imitates some functions of a PyTorch model.

Docs Updated

• docs/models/<model>/01-model-notes.md
• [ x ] docs/models/<model>/02-source-and-setup.md
• docs/datasets/candidates.md
• docs/datasets/selected.md
• Other:

What You Verified

Running the main ensemble (excluding using other models) will correctly start the mesonet server and produce a valid output. Stopping the server may fail on some errors, but handling this error may depend on how the facial_analyzer.py processing function is implemented and used in the future. The current solution is to kill the server process (see docs/models/<model>/02-source-and-setup.md Issues encountered + fixes).

Commands Run (if any)

Exact command usage and outputs provided in source-and-setup.md for MesoNet.

# Activate the ensemble virtual environment, if not already activated
source venv/bin/activate
# (venv) should appear before the prompt: "(venv) user/.../AI-Video-Detection$"
# Then in the repository directory, running the command will make a prediction on the given video
python3 -m backend.main --input-dir "./backend/dataset/FaceForensics++/original/002.mp4"

Screenshots Added?

• Yes — added to assets/ and referenced in markdown
• No — not applicable

Checklist

• [x ] Updated model docs OR dataset docs (as applicable)
• [ x ] Included links to sources (paper/repo/weights)
• [ x ] Included run commands + outputs (or screenshots)
• [ x ] Added screenshots to assets/ and referenced in markdown (if any)
• [ x ] No dataset files committed (verified .gitignore rules)