How akhildragon-07/deep_learning works
RMS-AV is a PyTorch research system for audio-visual deepfake detection, with audio-only, video-only, standard-fusion, and reliability-aware fusion models. Its main workflow prepares identity-safe data, trains models, evaluates clean and modality-missing cases, and serves predictions through an upload-oriented dashboard or standalone inference engine. A separate LAV-DF pipeline adds dataset inspection, matched preprocessing, training, evaluation, and inference. The graph follows documented capabilities and indexed imports; unsampled implementation details and undocumented runtime links are intentionally omitted.
Main entry points: dataset.py Pilot Dataset Loader · lavdf_pipeline/train.py LAV-DF Trainer · model.py RMS-AV and Baselines · train.py Pilot Trainer
Components 18
Data Preparation
- Pilot Dataset
teammates - Pilot Dataset Loader
dataset.py
RMS-AV Core
- RMS-AV and Baselines
model.py - Reliability Gate
reliability_gate.py - PG-ABN Model
model_pgabn.py
Training and Evaluation
- Pilot Trainer
train.py - PG-ABN Trainer
train_pgabn.py - Pilot Evaluator
evaluate.py
Prediction Serving
- Inference Engine
inference.py - Upload Dashboard
app.py
LAV-DF Pipeline
- LAV-DF Inspector
lavdf_pipeline/inspect_lavdf.py - LAV-DF Dataset
lavdf_pipeline/dataset.py - LAV-DF Preprocessing
lavdf_pipeline/preprocessing.py - LAV-DF Trainer
lavdf_pipeline/train.py - LAV-DF Evaluator
lavdf_pipeline/evaluate.py - LAV-DF Models
lavdf_pipeline/model.py - LAV-DF Inference
lavdf_pipeline/inference.py
Other components
- Researcher
Connections 28
- Researcher → Pilot Dataset Loader: prepares data
README.md - Pilot Dataset → Pilot Dataset Loader: loads samples
README.md - Pilot Dataset Loader → Pilot Trainer: supplies batches
train.py - Pilot Trainer → RMS-AV and Baselines: builds model
train.py - Pilot Trainer → Pilot Dataset Loader: loads dataset
train.py - RMS-AV and Baselines → Reliability Gate: uses gate
model.py - RMS-AV and Baselines → PG-ABN Model: imports model
model.py - Researcher → Pilot Trainer: trains models
README.md - Researcher → Pilot Evaluator: evaluates models
README.md - Pilot Evaluator → RMS-AV and Baselines: loads model
evaluate.py - Pilot Evaluator → Pilot Dataset Loader: loads dataset
evaluate.py - PG-ABN Trainer → PG-ABN Model: builds model
train_pgabn.py - PG-ABN Trainer → Pilot Dataset Loader: loads dataset
train_pgabn.py - Researcher → LAV-DF Inspector: audits dataset
- LAV-DF Inspector → LAV-DF Dataset: inspects annotations
- LAV-DF Dataset → LAV-DF Preprocessing: extracts modalities
lavdf_pipeline/dataset.py - LAV-DF Trainer → LAV-DF Dataset: loads dataset
lavdf_pipeline/train.py - LAV-DF Trainer → LAV-DF Preprocessing: uses preprocessing
lavdf_pipeline/train.py - LAV-DF Trainer → LAV-DF Models: builds model
lavdf_pipeline/train.py - Researcher → LAV-DF Trainer: trains models
lavdf_pipeline/train.py - LAV-DF Evaluator → LAV-DF Dataset: loads dataset
lavdf_pipeline/evaluate.py - LAV-DF Evaluator → LAV-DF Models: loads model
lavdf_pipeline/evaluate.py - Researcher → LAV-DF Evaluator: evaluates models
lavdf_pipeline/evaluate.py - Upload Dashboard → Inference Engine: requests predictions
app.py - Researcher → Upload Dashboard: uploads media
- Researcher → Inference Engine: submits media
- LAV-DF Inference → LAV-DF Models: loads model
lavdf_pipeline/inference.py - LAV-DF Inference → LAV-DF Preprocessing: preprocesses media
lavdf_pipeline/inference.py