# Data Diversity, Not Frequency Invariance: A Controlled and Self-Audited Study of Compression-Robust Deepfake Detection

arXiv: https://arxiv.org/abs/2608.28685 · HTML: https://mlsanity.com/paper/2608.28685/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 36 references that are in the map.
- First posted on arXiv 26 Aug 2026.

## Builds on (most cited first, 25 of 25)

- [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://mlsanity.com/paper/1905.11946/) (2019)
- [FiLM: Visual Reasoning with a General Conditioning Layer](https://mlsanity.com/paper/1709.07871/) (2017)
- [Xception: Deep Learning with Depthwise Separable Convolutions](https://mlsanity.com/paper/1610.02357/) (2016)
- [Unsupervised Domain Adaptation by Backpropagation](https://mlsanity.com/paper/1409.7495/) (2014)
- [FaceForensics++: Learning to Detect Manipulated Facial Images](https://mlsanity.com/paper/1901.08971/) (2019)
- [Celeb-DF: A Large-scale Challenging Dataset for DeepFake Forensics](https://mlsanity.com/paper/1909.12962/) (2019)
- [Thinking in Frequency: Face Forgery Detection by Mining Frequency-aware Clues](https://mlsanity.com/paper/2007.09355/) (2020)
- [Leveraging Frequency Analysis for Deep Fake Image Recognition](https://mlsanity.com/paper/2003.08685/) (2020)
- [Towards Universal Fake Image Detectors that Generalize Across Generative Models](https://mlsanity.com/paper/2302.10174/) (2023)
- [Detecting Deepfakes with Self-Blended Images](https://mlsanity.com/paper/2204.08376/) (2022)
- [Lips Don't Lie: A Generalisable and Robust Approach to Face Forgery Detection](https://mlsanity.com/paper/2012.07657/) (2020)
- [Watch your Up-Convolution: CNN Based Generative Deep Neural Networks are Failing to Reproduce Spectral Distributions](https://mlsanity.com/paper/2003.01826/) (2020)
- [Generalizing Face Forgery Detection with High-frequency Features](https://mlsanity.com/paper/2103.12376/) (2021)
- [Spatial-Phase Shallow Learning: Rethinking Face Forgery Detection in Frequency Domain](https://mlsanity.com/paper/2103.01856/) (2021)
- [WildDeepfake: A Challenging Real-World Dataset for Deepfake Detection](https://mlsanity.com/paper/2101.01456/) (2021)
- [Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection](https://mlsanity.com/paper/2312.16649/) (2023)
- [DeepfakeBench: A Comprehensive Benchmark of Deepfake Detection](https://mlsanity.com/paper/2307.01426/) (2023)
- [Leveraging Real Talking Faces via Self-Supervision for Robust Forgery Detection](https://mlsanity.com/paper/2201.07131/) (2022)
- [Weak baselines and reporting biases lead to overoptimism in machine learning for fluid-related partial differential equations](https://mlsanity.com/paper/2407.07218/) (2024)
- [Sample and Computation Redistribution for Efficient Face Detection](https://mlsanity.com/paper/2105.04714/) (2021)
- [Pre-registration for Predictive Modeling](https://mlsanity.com/paper/2311.18807/) (2023)
- [Metric Learning for Anti-Compression Facial Forgery Detection](https://mlsanity.com/paper/2103.08397/) (2021)
- [Position: Embracing Negative Results in Machine Learning](https://mlsanity.com/paper/2406.03980/) (2024)
- [Sensitivity Decouple Learning for Image Compression Artifacts Reduction](https://mlsanity.com/paper/2405.09291/) (2024)
- [Pay Less Attention to Deceptive Artifacts: Robust Detection of Compressed Deepfakes on Online Social Networks](https://mlsanity.com/paper/2506.20548/) (2025)

## Abstract (arXiv)

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.