analiza aiqbench
This study challenges the prevailing assumption that frequency-domain features are essential for compression-robust deepfake detection, demonstrating instead that data diversity from real-world codecs outperforms complex architectural tricks like adversarial branches. By rigorously controlling for capacity and augmentation biases, the research proves that robustness stems primarily from training on varied video qualities rather than specialized frequency-invariant learning mechanisms.
Data Diversity, Not Frequency Invariance: A Controlled and Self-Audited Study of Compression-Robust Deepfake Detection
arXiv:2608.28685v1 Announce Type: cross Abstract: Frequency features and compression-invariant representation learning are widely assumed to be key to deepfake detection that survives video compression. We test this with CAFRL - block-DCT and FFT-phase streams, compression-level-conditioned band attention, and adversarial (gradient-reversal) compression invariance - and report a controlled negative. Under a pre-registered protocol with capacity- and augmentation-matched controls, a plain EfficientNet-B0 on multi-quality data beat CAFRL as specified at every compression level on the FaceForensics++ test split, by 3.66 AUC points at CRF 40 (paired, single seed). A self-audit of our own negative found four defects biased against the frequency hypothesis, and pre-specified re-tests repairing all four showed the deficit to be a recipe artifact, not an architecture failure: the baseline recipe recovered 3.96 points over the matching shipped-recipe variant. The frequency path made no detectable difference: discriminative alone (standalone validation AUC 0.91-0.98 late in training) but of no marginal value under this fusion, at two feature widths of one 4.0 M trunk, every seed-pooled interval for the intra-dataset compression contrasts including zero; on the single held-out manipulation tested, the fair variants sat below the plain backbone. The adversarial branch, as specified, added nothing and degraded its own conditioning estimator; at the fair recipe it is untested. Robustness under single-pass H.264 re-encoding came instead from data diversity: real constant-rate-factor variants beat synthetic JPEG augmentation by 7.3 points (single runs, non-overlapping intervals). The evidence is FaceForensics++-family, GAN-era and single-codec. Match controls on training recipe as well as capacity, and buy compression robustness with codec diversity before architecture.
Abstract:Frequency features and compression-invariant representation learning are widely assumed to be key to deepfake detection that survives video compression. We test this with CAFRL - block-DCT and FFT-phase streams, compression-level-conditioned band attention, and adversarial (gradient-reversal) compression invariance - and report a controlled negative. Under a pre-registered protocol with capacity- and augmentation-matched controls, a plain EfficientNet-B0 on multi-quality data beat CAFRL as specified at every compression level on the FaceForensics++ test split, by 3.66 AUC points at CRF 40 (paired, single seed). A self-audit of our own negative found four defects biased against the frequency hypothesis, and pre-specified re-tests repairing all four showed the deficit to be a recipe artifact, not an architecture failure: the baseline recipe recovered 3.96 points over the matching shipped-recipe variant. The frequency path made no detectable difference: discriminative alone (standalone validation AUC 0.91-0.98 late in training) but of no marginal value under this fusion, at two feature widths of one 4.0 M trunk, every seed-pooled interval for the intra-dataset compression contrasts including zero; on the single held-out manipulation tested, the fair variants sat below the plain backbone. The adversarial branch, as specified, added nothing and degraded its own conditioning estimator; at the fair recipe it is untested. Robustness under single-pass H.264 re-encoding came instead from data diversity: real constant-rate-factor variants beat synthetic JPEG augmentation by 7.3 points (single runs, non-overlapping intervals). The evidence is FaceForensics++-family, GAN-era and single-codec. Match controls on training recipe as well as capacity, and buy compression robustness with codec diversity before architecture.
38 pages, 4 figures. Submitted to IEEE Access. Code, per-video predictions and a claims-to-artifacts map: this https URL (v1.0.0); archival deposit: this https URL
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Multimedia (cs.MM)
arXiv:2608.28685 [cs.CV]
(or arXiv:2608.28685v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2608.28685 arXiv-issued DOI via DataCite (pending registration)
Text integral sub licența deschisă a publicației · 300 cuvinte
Subiecte conexe
Auditing Pairwise Equivalence Judgments: Self-Critique Effects and Diversity Measurement in Multi-Agent Hypothesis Generation
arXiv · 7 oct. 2026
I Did This Without Your Knowledge
LessWrong · 17 oct. 2026
Biomedical Imaging's Real Bottleneck Is the Data, Not the Model
Databricks · 8 oct. 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · 17 sept. 2026