- DAVID-Net: 160-Run Experimental Archive and Benchmark Telemetry
DAVID-Net: 160-Run Experimental Archive and Benchmark Telemetry
Official experimental archive, multi-seed checkpoints, training event streams, and evaluation telemetry for DAVID-Net (Decoupled Audio-Visual Forensic Network).
This repository contains the persistent, uncompressed experimental record for the entire 160-run / 5-seed empirical campaign (135 core campaign runs across 27 experimental configurations + 25 auxiliary baseline, scaling, and recovery runs, seeds $\mathcal{S} \in {7, 42, 123, 456, 2024}$).
- Source Code Repository: GitHub: MIHMahmudEli/david-net-av
- Core Pretrained Model: Hugging Face Models: MIHMahmudEli/david-net-av
- Preprocessed Datasets: Hugging Face Datasets: MIHMahmudEli/david-net-av
- Live Interactive Demo: Vercel: david-net-av.vercel.app
- License: MIT License
- Format: SafeTensors (
best_model/model.safetensorsper run), JSON configurations, CSV telemetry
π Experimental Campaign Overview
The empirical validation of DAVID-Net reflects a comprehensive matrix of 160 multi-seed training runs partitioned across five primary experimental stages:
- Stage-0: Quadrant-Aware Contrastive Pretraining (QACP) (30 runs):
- 6 self-supervised pretraining variants $\times$ 5 random seeds (seeds 7, 42, 123, 456, 2024).
- Generates proxy synthetic media from pristine recordings using self-blended video and vocoder copy-synthesis speech.
- Stage-1: In-Domain Baselines & Proposed Architecture (35 runs):
- 7 model architectures $\times$ 5 random seeds: Video Probe, Audio Probe, AASIST, EfficientNet-B4, Late Fusion, DAVID-Net Phase A, and DAVID-Net End-to-End Phase B.
- Evaluated under strict identity-disjoint FakeAVCeleb partitioning (0.0% train/test subject identity overlap).
- Systematic Component Ablation Study (55 runs):
- 11 ablation variants $\times$ 5 random seeds: isolating cross-modal synchronization (
no_sync), subspace disentanglement (no_disentangle), localization heads (no_loc), modality dropout (no_moddrop), single-task loss (single_task), pretraining (no_qacp), and sub-component pretraining objectives.
- 11 ablation variants $\times$ 5 random seeds: isolating cross-modal synchronization (
- Leave-One-Generator-Out (LOGO) Study (15 runs):
- 3 withheld generator families (
logo_wav2lip,logo_fsgan,logo_faceswap) $\times$ 5 random seeds to measure zero-shot cross-generator generalization.
- 3 withheld generator families (
- Auxiliary Baseline & Scaling Campaign (25 runs):
- Additional baseline explorations (AVFF, AVoiD-DF, Cross-Attention) and scaled data sweeps across all 5 seeds.
π¬ In-Domain Benchmark Summary ($N=5$ Seeds)
All 25 evaluated model configurations averaged over five independent seeds ($\mathcal{S} \in {7, 42, 123, 456, 2024}$):
| Model / Configuration | Seeds | Video AUC | Audio AUC | Clip AUC |
|---|---|---|---|---|
aasist |
5 | β | 1.000 Β± 0.000 | 0.748 Β± 0.004 |
audio-probe |
5 | β | 0.999 Β± 0.001 | 0.732 Β± 0.001 |
avff |
5 | 0.978 Β± 0.003 | 1.000 Β± 0.000 | 0.981 Β± 0.003 |
avoid-df |
5 | 0.922 Β± 0.003 | 1.000 Β± 0.000 | 0.938 Β± 0.004 |
cross-attn |
5 | 0.938 Β± 0.007 | 0.999 Β± 0.001 | 0.939 Β± 0.008 |
davidnet |
5 | 0.979 Β± 0.003 | 0.999 Β± 0.002 | 0.976 Β± 0.006 |
davidnet-compose_quad |
5 | 0.979 Β± 0.004 | 1.000 Β± 0.000 | 0.978 Β± 0.003 |
davidnet-e2e |
5 | 0.976 Β± 0.005 | 1.000 Β± 0.000 | 0.978 Β± 0.005 |
davidnet-legacy |
5 | 0.985 Β± 0.004 | 1.000 Β± 0.000 | 0.985 Β± 0.002 |
davidnet-logo_faceswap |
5 | 0.912 Β± 0.006 | 1.000 Β± 0.000 | 0.903 Β± 0.003 |
davidnet-logo_fsgan |
5 | 0.934 Β± 0.004 | 1.000 Β± 0.000 | 0.932 Β± 0.007 |
davidnet-logo_wav2lip |
5 | 0.425 Β± 0.018 | 0.994 Β± 0.010 | 0.919 Β± 0.014 |
davidnet-no_disentangle |
5 | 0.979 Β± 0.002 | 1.000 Β± 0.000 | 0.979 Β± 0.001 |
davidnet-no_loc |
5 | 0.980 Β± 0.004 | 1.000 Β± 0.000 | 0.981 Β± 0.004 |
davidnet-no_moddrop |
5 | 0.980 Β± 0.003 | 1.000 Β± 0.000 | 0.981 Β± 0.004 |
davidnet-no_qacp |
5 | 0.981 Β± 0.002 | 1.000 Β± 0.000 | 0.983 Β± 0.003 |
davidnet-no_sync |
5 | 0.973 Β± 0.005 | 1.000 Β± 0.000 | 0.975 Β± 0.011 |
davidnet-qacp_no_copysynth |
5 | 0.980 Β± 0.003 | 0.999 Β± 0.001 | 0.979 Β± 0.006 |
davidnet-qacp_no_mismatch |
5 | 0.976 Β± 0.003 | 1.000 Β± 0.000 | 0.977 Β± 0.003 |
davidnet-qacp_no_selfblend |
5 | 0.977 Β± 0.006 | 1.000 Β± 0.001 | 0.978 Β± 0.006 |
davidnet-qacp_scaled |
5 | 0.979 Β± 0.002 | 1.000 Β± 0.000 | 0.978 Β± 0.006 |
davidnet-single_task |
5 | 0.978 Β± 0.003 | 1.000 Β± 0.000 | 0.978 Β± 0.005 |
effnet-b4 |
5 | 0.996 Β± 0.001 | β | 0.926 Β± 0.002 |
late-fusion |
5 | 0.914 Β± 0.020 | 1.000 Β± 0.000 | 0.918 Β± 0.015 |
video-probe |
5 | 0.896 Β± 0.008 | β | 0.830 Β± 0.007 |
Key Diagnostic Findings
- Decoupled Attribution: Multi-task learning does not handicap individual modality branches: DAVID-Net achieves $0.979 \pm 0.003$ Video AUC and $0.999 \pm 0.002$ Audio AUC, while reaching $0.976 \pm 0.006$ Clip AUC, outperforming monolithic Late Fusion ($0.918 \pm 0.015$).
- Cross-Generator Floor (LOGO): Maintains $\ge 90.3%$ Clip AUC across all withheld generator families under leave-one-generator-out evaluation.
- Empirical Calibration: Attains low Expected Calibration Error (ECE, 15 bins) of $0.032$ (visual) and $0.024$ (acoustic) on the test split.
- Zero Cross-Stream Confusion: Analysis of the normalized confusion matrix reveals 0.0% confusion between opposite single-stream manipulations (RVFA vs. FVRA).
π Repository Organization & File Hierarchy
The repository is structured strictly according to publication-grade open-science standards:
MIHMahmudEli/davidnet-experiments/
βββ data/
β βββ SPLITS_SHA256.json # Cryptographic SHA-256 hashes of all partitions
β βββ feature_set.json # Preprocessing & feature extraction parameters
β βββ manifests/ # Manifest JSONL indices for 7 benchmark datasets
β βββ splits/ # Identity-disjoint strict & LOGO train/val/test splits
βββ experiments/
β βββ EXP_{ID}_{NAME}_s{SEED}/
β βββ best_model/
β β βββ model.safetensors # Pinned model weights for this seed (SafeTensors)
β β βββ config.json # Architecture parameters
β β βββ meta.json # Validation metrics & checkpoint metadata
β βββ configs/
β β βββ config.json # Full hyperparameter configuration
β β βββ environment.json # Hardware, OS, and PyTorch runtime telemetry
β βββ metrics/
β β βββ training_history.csv # Per-epoch loss, learning rates, and validation metrics
β β βββ test_metrics.json # Final test evaluation metrics
β β βββ confusion_matrix.csv # 4-quadrant assignment counts
β β βββ classification_report.csv
β βββ figures/ # Per-run ROC, PR, calibration, and distribution plots
βββ peer_review_audit/
β βββ profiling/ # Nvidia T4 runtime latency, memory, and FLOPs profiling
β βββ stats/ # DeLong test $z$-scores, $p$-values, and bootstrap 95% CIs
βββ registry/
β βββ experiments.json # Master campaign tracking manifest (hashes, epochs, status)
βββ assets/
β βββ fig_davidnet.png # High-resolution architectural overview
βββ .gitattributes # Git LFS configuration
βββ README.md # This Model Card & benchmark leaderboard
π» Programmatic Usage: Loading Run Artifacts & Weights
You can inspect and load the weights, training logs, or configs of any run programmatically using huggingface_hub and safetensors:
import pandas as pd
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
repo_id = "MIHMahmudEli/davidnet-experiments"
# 1. Download training history for EXP_011 (DAVID-Net Phase A, Seed 42)
history_path = hf_hub_download(
repo_id=repo_id,
filename="experiments/EXP_011_davidnet_s42/metrics/training_history.csv"
)
df_history = pd.read_csv(history_path)
print("Training Epochs:", len(df_history))
# 2. Download checkpoint weights (SafeTensors)
weights_path = hf_hub_download(
repo_id=repo_id,
filename="experiments/EXP_011_davidnet_s42/best_model/model.safetensors"
)
state_dict = load_file(weights_path)
print(f"Loaded {len(state_dict)} tensor keys successfully.")
π Complete 160-Run Campaign Registry
| ID | Configuration | Seed | Status | Test Clip AUC |
|---|---|---|---|---|
| EXP_001 | video-probe |
42 | completed | 0.8373 |
| EXP_002 | audio-probe |
42 | completed | 0.7294 |
| EXP_003 | late-fusion |
42 | completed | 0.9307 |
| EXP_004 | aasist |
42 | completed | 0.7471 |
| EXP_005 | effnet-b4 |
42 | completed | 0.9232 |
| EXP_006 | qacp |
42 | completed | β |
| EXP_007 | qacp-no_sync |
42 | completed | β |
| EXP_008 | qacp-no_mismatch |
42 | completed | β |
| EXP_009 | qacp-no_copysynth |
42 | completed | β |
| EXP_010 | qacp-no_selfblend |
42 | completed | β |
| EXP_011 | davidnet |
42 | completed | 0.9725 |
| EXP_012 | davidnet-no_qacp |
42 | completed | 0.9857 |
| EXP_013 | davidnet-no_sync |
42 | completed | 0.9810 |
| EXP_014 | davidnet-no_disentangle |
42 | completed | 0.9802 |
| EXP_015 | davidnet-no_loc |
42 | completed | 0.9799 |
| EXP_016 | davidnet-single_task |
42 | completed | 0.9747 |
| EXP_017 | davidnet-no_moddrop |
42 | completed | 0.9762 |
| EXP_018 | davidnet-compose_quad |
42 | completed | 0.9766 |
| EXP_019 | davidnet-qacp_no_mismatch |
42 | completed | 0.9796 |
| EXP_020 | davidnet-qacp_no_copysynth |
42 | completed | 0.9701 |
| EXP_021 | davidnet-qacp_no_selfblend |
42 | completed | 0.9708 |
| EXP_022 | davidnet-logo_wav2lip |
42 | completed | 0.9274 |
| EXP_023 | davidnet-logo_fsgan |
42 | completed | 0.9350 |
| EXP_024 | davidnet-logo_faceswap |
42 | completed | 0.9012 |
| EXP_025 | qacp-legacy |
42 | completed | β |
| EXP_026 | davidnet-legacy |
42 | completed | 0.9826 |
| EXP_027 | davidnet-e2e |
42 | completed | 0.9793 |
| EXP_028 | video-probe |
123 | completed | 0.8349 |
| EXP_029 | audio-probe |
123 | completed | 0.7336 |
| EXP_030 | late-fusion |
123 | completed | 0.9273 |
| EXP_031 | aasist |
123 | completed | 0.7535 |
| EXP_032 | effnet-b4 |
123 | completed | 0.9259 |
| EXP_033 | qacp |
123 | completed | β |
| EXP_034 | qacp-no_sync |
123 | completed | β |
| EXP_035 | qacp-no_mismatch |
123 | completed | β |
| EXP_036 | qacp-no_copysynth |
123 | completed | β |
| EXP_037 | qacp-no_selfblend |
123 | completed | β |
| EXP_038 | davidnet |
123 | completed | 0.9829 |
| EXP_039 | davidnet-no_qacp |
123 | completed | 0.9797 |
| EXP_040 | davidnet-no_sync |
123 | completed | 0.9578 |
| EXP_041 | davidnet-no_disentangle |
123 | completed | 0.9799 |
| EXP_042 | davidnet-no_loc |
123 | completed | 0.9768 |
| EXP_043 | davidnet-single_task |
123 | completed | 0.9779 |
| EXP_044 | davidnet-no_moddrop |
123 | completed | 0.9803 |
| EXP_045 | davidnet-compose_quad |
123 | completed | 0.9802 |
| EXP_046 | davidnet-qacp_no_mismatch |
123 | completed | 0.9739 |
| EXP_047 | davidnet-qacp_no_copysynth |
123 | completed | 0.9752 |
| EXP_048 | davidnet-qacp_no_selfblend |
123 | completed | 0.9770 |
| EXP_049 | davidnet-logo_wav2lip |
123 | completed | 0.8979 |
| EXP_050 | davidnet-logo_fsgan |
123 | completed | 0.9330 |
| EXP_051 | davidnet-logo_faceswap |
123 | completed | 0.9083 |
| EXP_052 | qacp-legacy |
123 | completed | β |
| EXP_053 | davidnet-legacy |
123 | completed | 0.9865 |
| EXP_054 | davidnet-e2e |
123 | completed | 0.9723 |
| EXP_055 | video-probe |
456 | completed | 0.8204 |
| EXP_056 | audio-probe |
456 | completed | 0.7319 |
| EXP_057 | late-fusion |
456 | completed | 0.9004 |
| EXP_058 | aasist |
456 | completed | 0.7447 |
| EXP_059 | effnet-b4 |
456 | completed | 0.9274 |
| EXP_060 | qacp |
456 | completed | β |
| EXP_061 | qacp-no_sync |
456 | completed | β |
| EXP_062 | qacp-no_mismatch |
456 | completed | β |
| EXP_063 | qacp-no_copysynth |
456 | completed | β |
| EXP_064 | qacp-no_selfblend |
456 | completed | β |
| EXP_065 | davidnet |
456 | completed | 0.9726 |
| EXP_066 | davidnet-no_qacp |
456 | completed | 0.9828 |
| EXP_067 | davidnet-no_sync |
456 | completed | 0.9814 |
| EXP_068 | davidnet-no_disentangle |
456 | completed | 0.9776 |
| EXP_069 | davidnet-no_loc |
456 | completed | 0.9858 |
| EXP_070 | davidnet-single_task |
456 | completed | 0.9819 |
| EXP_071 | davidnet-no_moddrop |
456 | completed | 0.9855 |
| EXP_072 | davidnet-compose_quad |
456 | completed | 0.9828 |
| EXP_073 | davidnet-qacp_no_mismatch |
456 | completed | 0.9802 |
| EXP_074 | davidnet-qacp_no_copysynth |
456 | completed | 0.9833 |
| EXP_075 | davidnet-qacp_no_selfblend |
456 | completed | 0.9840 |
| EXP_076 | davidnet-logo_wav2lip |
456 | completed | 0.9328 |
| EXP_077 | davidnet-logo_fsgan |
456 | completed | 0.9316 |
| EXP_078 | davidnet-logo_faceswap |
456 | completed | 0.9054 |
| EXP_079 | qacp-legacy |
456 | completed | β |
| EXP_080 | davidnet-legacy |
456 | completed | 0.9877 |
| EXP_081 | davidnet-e2e |
456 | completed | 0.9748 |
| EXP_082 | audio-probe |
7 | completed | 0.7315 |
| EXP_083 | qacp-no_selfblend |
7 | completed | β |
| EXP_084 | late-fusion |
2024 | completed | 0.9032 |
| EXP_085 | qacp-no_copysynth |
2024 | completed | β |
| EXP_086 | davidnet-no_qacp |
2024 | completed | 0.9806 |
| EXP_087 | qacp-no_mismatch |
2024 | completed | β |
| EXP_088 | qacp-no_copysynth |
7 | completed | β |
| EXP_089 | qacp |
2024 | completed | β |
| EXP_090 | video-probe |
2024 | completed | 0.8260 |
| EXP_091 | qacp-no_selfblend |
2024 | completed | β |
| EXP_092 | video-probe |
7 | completed | 0.8334 |
| EXP_093 | late-fusion |
7 | completed | 0.9302 |
| EXP_094 | aasist |
2024 | completed | 0.7455 |
| EXP_095 | qacp-no_sync |
2024 | completed | β |
| EXP_096 | qacp-legacy |
2024 | completed | β |
| EXP_097 | davidnet-qacp_no_mismatch |
2024 | completed | 0.9761 |
| EXP_098 | davidnet-logo_wav2lip |
2024 | completed | 0.9096 |
| EXP_099 | davidnet-qacp_no_selfblend |
7 | completed | 0.9742 |
| EXP_100 | qacp |
7 | completed | β |
| EXP_101 | qacp-no_sync |
7 | completed | β |
| EXP_102 | qacp-no_mismatch |
7 | completed | β |
| EXP_103 | qacp-legacy |
7 | completed | β |
| EXP_104 | effnet-b4 |
2024 | completed | 0.9279 |
| EXP_105 | davidnet-legacy |
2024 | completed | 0.9831 |
| EXP_106 | davidnet-compose_quad |
7 | completed | 0.9740 |
| EXP_107 | davidnet-logo_fsgan |
7 | completed | 0.9401 |
| EXP_108 | davidnet-no_moddrop |
7 | completed | 0.9782 |
| EXP_109 | audio-probe |
2024 | completed | 0.7313 |
| EXP_110 | davidnet-logo_wav2lip |
7 | completed | 0.9256 |
| EXP_111 | davidnet-qacp_no_mismatch |
7 | completed | 0.9750 |
| EXP_112 | davidnet-no_loc |
7 | completed | 0.9804 |
| EXP_113 | davidnet-qacp_no_selfblend |
2024 | completed | 0.9837 |
| EXP_114 | davidnet |
2024 | completed | 0.9694 |
| EXP_115 | davidnet-no_disentangle |
2024 | completed | 0.9799 |
| EXP_116 | davidnet-no_moddrop |
2024 | completed | 0.9841 |
| EXP_117 | davidnet-no_sync |
2024 | completed | 0.9826 |
| EXP_118 | aasist |
7 | completed | 0.7507 |
| EXP_119 | davidnet-no_disentangle |
7 | completed | 0.9778 |
| EXP_120 | davidnet-no_sync |
7 | completed | 0.9700 |
| EXP_121 | davidnet-legacy |
7 | completed | 0.9831 |
| EXP_122 | davidnet-qacp_no_copysynth |
7 | completed | 0.9858 |
| EXP_123 | davidnet-compose_quad |
2024 | completed | 0.9770 |
| EXP_124 | davidnet-no_qacp |
7 | completed | 0.9852 |
| EXP_125 | davidnet-qacp_no_copysynth |
2024 | completed | 0.9795 |
| EXP_126 | davidnet-logo_fsgan |
2024 | completed | 0.9220 |
| EXP_127 | davidnet-no_loc |
2024 | completed | 0.9838 |
| EXP_128 | davidnet-single_task |
2024 | completed | 0.9841 |
| EXP_129 | davidnet-e2e |
2024 | completed | 0.9845 |
| EXP_130 | davidnet-logo_faceswap |
7 | completed | 0.9010 |
| EXP_131 | davidnet-single_task |
7 | completed | 0.9730 |
| EXP_132 | effnet-b4 |
7 | completed | 0.9268 |
| EXP_133 | davidnet |
7 | completed | 0.9807 |
| EXP_134 | davidnet-e2e |
7 | completed | 0.9797 |
| EXP_135 | davidnet-logo_faceswap |
2024 | completed | 0.9008 |
| EXP_136 | avoid-df |
42 | completed | 0.9400 |
| EXP_137 | avoid-df |
2024 | completed | 0.9313 |
| EXP_138 | avoid-df |
123 | completed | 0.9376 |
| EXP_139 | avoid-df |
456 | completed | 0.9422 |
| EXP_140 | cross-attn |
123 | completed | 0.9346 |
| EXP_141 | cross-attn |
2024 | completed | 0.9320 |
| EXP_142 | avoid-df |
7 | completed | 0.9376 |
| EXP_143 | cross-attn |
456 | completed | 0.9377 |
| EXP_144 | cross-attn |
7 | completed | 0.9371 |
| EXP_145 | cross-attn |
42 | completed | 0.9523 |
| EXP_146 | avff |
123 | completed | 0.9797 |
| EXP_147 | avff |
42 | completed | 0.9841 |
| EXP_148 | avff |
2024 | completed | 0.9846 |
| EXP_149 | avff |
7 | completed | 0.9816 |
| EXP_150 | avff |
456 | completed | 0.9769 |
| EXP_151 | qacp-scaled |
7 | completed | β |
| EXP_152 | qacp-scaled |
123 | completed | β |
| EXP_153 | qacp-scaled |
456 | completed | β |
| EXP_154 | qacp-scaled |
42 | completed | β |
| EXP_155 | qacp-scaled |
2024 | completed | β |
| EXP_156 | davidnet-qacp_scaled |
7 | completed | 0.9823 |
| EXP_157 | davidnet-qacp_scaled |
123 | completed | 0.9804 |
| EXP_158 | davidnet-qacp_scaled |
42 | completed | 0.9816 |
| EXP_159 | davidnet-qacp_scaled |
456 | completed | 0.9796 |
| EXP_160 | davidnet-qacp_scaled |
2024 | completed | 0.9682 |
βοΈ Ethics & Responsible AI Statement
All experimental evaluations in this repository were conducted under strict academic research ethics. Seven benchmark corpora were acquired under their respective research end-user licensing agreements; no raw biometric media is redistributed. The system is strictly defensive by design and contains no generative synthesis capability. Model outputs provide posterior probability calibration signals to assist human forensics experts, not autonomous legal determinations.
π Citation
@article{hossain2026davidnet,
title={DAVID-Net: Disentangled Audio-Visual Forensic Attribution under Identity-Disjoint Evaluation},
author={Hossain, Moshin Ibna and Khan, Alif Hasan and Shakib, Shadman and Jam, MD. Yeahyea and Hasan, Mohammad Mahmudul},
journal={Machine Learning with Applications},
year={2026},
note={Under review}
}
@thesis{hossain2026thesis,
title={Decoupled Multimodal Forensic Attribution and Desynchronization Detection},
author={Hossain, Moshin Ibna and Khan, Alif Hasan and Shakib, Shadman and Jam, MD. Yeahyea},
school={American International University-Bangladesh},
year={2026}
}
