Add deepfake_detection core module

.pre-commit-config.yaml

@@ -59,12 +59,11 @@ repos:
     hooks:
       - id: pytest
         name: pytest
-        entry: pytest -m "not integration_test"
-        language: python
+        entry: uv run pytest -m "not integration_test"
+        language: system
         types: [python]
         pass_filenames: false
         always_run: true
-        additional_dependencies: [pytest]
 
 ci:
     autofix_commit_msg: |

README.md

@@ -37,6 +37,7 @@ vector-aixpert/
 │
 ├── src/aixpert/
 │   ├── controlled_images/                  # Baseline vs fairness-aware image generation
+│   ├── deepfake_detection/                 # Curated multimodal deepfake data preparation
 │   ├── data_generation/
 │   │   ├── synthetic_data_generation/
 │   │   │   ├── images/                     # Domain/risk-specific image + VQA generation
@@ -94,6 +95,9 @@ uv run mkdocs serve
 - **Controlled Images** — Matched baseline vs fairness-aware images across professions.
   ➜ [`src/aixpert/controlled_images/README.md`](src/aixpert/controlled_images/README.md)
 
+- **Deepfake Detection** — FACT-HO bundle preparation for LAV-DF, FakeAVCeleb, and VCapAV.
+  ➜ [`src/aixpert/deepfake_detection/README.md`](src/aixpert/deepfake_detection/README.md)
+
 - **Agent Pipeline (CrewAI)** — Single-agent orchestration for prompt/image/metadata generation.
   ➜ [`src/aixpert/data_generation/agent_pipeline/README.md`](src/aixpert/data_generation/agent_pipeline/README.md)
 
@@ -149,4 +153,4 @@ Resources used in preparing this research were provided, in part, by the Provinc
 
 This work is part of the AIXpert project, funded by the **European Union's Horizon Europe Research and Innovation Programme** under Grant Agreement No. **101214389**, and the **Swiss State Secretariat for Education, Research and Innovation (SERI)**. Views expressed are those of the authors and do not necessarily reflect those of the European Union or funding authorities.
 
-🌐 [Project Website](https://aixpert-project.eu/) · [LinkedIn](https://www.linkedin.com/company/aixpert-project/) · [X/Twitter](https://x.com/AIXPERT_project) · [YouTube](https://www.youtube.com/@AIXPERT_project)
+🌐 [Project Website](https://aixpert-project.eu/) · [LinkedIn](https://www.linkedin.com/company/aixpert-project/) · [X/Twitter](https://x.com/AIXPERT_project) · [YouTube](https://www.youtube.com/@AIXPERT_project)

_typos.toml

@@ -11,3 +11,4 @@ LLM = "LLM"  # Large Language Model
 LLMs = "LLMs"  # Large Language Models (plural)
 VQA = "VQA"  # Visual Question Answering
 IG = "IG"  # Integrated Gradients
+HumAIne = "HumAIne"  # EU project name

docs/projects.md

@@ -78,4 +78,4 @@ Statistical metrics (e.g. Statistical Parity, Equal Opportunity), zero-shot expl
 
 - See [CONTRIBUTING.md](https://github.com/VectorInstitute/vector-aixpert/blob/main/CONTRIBUTING.md) for coding standards (PEP8, Google docstrings), pre-commit hooks (`ruff`, `mypy`, `typos`, `nbQA`), branching, and tests.
 - **Run docs locally:** `uv sync --no-group docs` then `mkdocs serve` → [http://127.0.0.1:8000](http://127.0.0.1:8000)
-- **CI:** GitHub Actions (`code_checks.yml`, `unit_tests.yml`, `integration_tests.yml`)
+- **CI:** GitHub Actions (`code_checks.yml`, `unit_tests.yml`, `integration_tests.yml`)

pyproject.toml

@@ -132,6 +132,7 @@ ignore = [
 # Ignore import violations in all `__init__.py` files.
 [tool.ruff.lint.per-file-ignores]
 "__init__.py" = ["E402", "F401", "F403", "F811"]
+"src/aixpert/deepfake_detection/builders.py" = ["PLR0912"]
 # Ignoring undocumented public functions, public init, magic method, and magic numbers in tests folder
 "tests/*" = ["D103", "D105", "D107"]
 

src/aixpert/deepfake_detection/README.md

@@ -0,0 +1,38 @@
+# Deepfake Detection
+
+This module packages the most reviewable and reusable parts of the current
+multimodal deepfake work into the `vector-aixpert` monorepo.
+
+## Scope
+
+The first version focuses on data preparation rather than full training:
+
+- FACT-HO sample and bundle domain objects
+- deterministic grouping and split helpers
+- dataset builders for `LAV-DF`, `FakeAVCeleb`, and manifest-first `VCapAV`
+- a small CLI for one-sample smoke summaries
+
+## Why this is curated
+
+The original working directory contains many experiment scripts, environment
+fixes, and cluster-specific launchers. For a first monorepo integration, this
+module keeps only the parts that are easiest to review, test, and scale.
+
+That means this initial contribution intentionally excludes:
+
+- repeated training variants
+- plotting and monitoring helpers
+- local outputs and checkpoints
+- user-specific absolute paths
+
+## Example
+
+From the repository root:
+
+```bash
+uv run python -m aixpert.deepfake_detection.cli summarize \
+  --dataset vcapav \
+  --data-root /path/to/data \
+  --metadata-path /path/to/vcapav_manifest.jsonl \
+  --vcapav-split-strategy metadata
+```

src/aixpert/deepfake_detection/__init__.py

@@ -0,0 +1,36 @@
+"""Curated utilities for multimodal deepfake dataset preparation."""
+
+from aixpert.deepfake_detection.builders import (
+    DatasetPartitions,
+    FakeAVCelebBuilder,
+    FakeAVCelebConfig,
+    LAVDFBuilder,
+    LAVDFConfig,
+    SelectionLimits,
+    SplitConfig,
+    VCapAVBuilder,
+    VCapAVConfig,
+)
+from aixpert.deepfake_detection.core import (
+    FactHOBundle,
+    FactHOSample,
+    assign_group_indices,
+    build_bundles_from_samples,
+)
+
+
+__all__ = [
+    "DatasetPartitions",
+    "FactHOBundle",
+    "FactHOSample",
+    "FakeAVCelebBuilder",
+    "FakeAVCelebConfig",
+    "LAVDFBuilder",
+    "LAVDFConfig",
+    "SelectionLimits",
+    "SplitConfig",
+    "VCapAVBuilder",
+    "VCapAVConfig",
+    "assign_group_indices",
+    "build_bundles_from_samples",
+]