Ekta Prashnani

Ekta Prashnani, Michael Goebel, B. S. Manjunath

We propose PhaseForensics, a DeepFake (DF) video detection method that leverages a phase-based motion representation of facial temporal dynamics. Existing methods relying on temporal inconsistencies for DF detection present many advantages over the typical frame-based methods. However, they still show limited cross-dataset generalization and robustness to common distortions. These shortcomings are partially due to error-prone motion estimation and landmark tracking, or the susceptibility of the pixel intensity-based features to spatial distortions and the cross-dataset domain shifts. Our key insight to overcome these issues is to leverage the temporal phase variations in the band-pass components of the Complex Steerable Pyramid on face sub-regions. This not only enables a robust estimate of the temporal dynamics in these regions, but is also less prone to cross-dataset variations. Furthermore, the band-pass filters used to compute the local per-frame phase form an effective defense against the perturbations commonly seen in gradient-based adversarial attacks. Overall, with PhaseForensics, we show improved distortion and adversarial robustness, and state-of-the-art cross-dataset generalization, with 91.2% video-level AUC on the challenging CelebDFv2 (a recent state-of-the-art compares at 86.9%).

Satish Kumar, ASM Iftekhar, Ekta Prashnani et al.

This paper describes LOCL (Learning Object Attribute Composition using Localization) that generalizes composition zero shot learning to objects in cluttered and more realistic settings. The problem of unseen Object Attribute (OA) associations has been well studied in the field, however, the performance of existing methods is limited in challenging scenes. In this context, our key contribution is a modular approach to localizing objects and attributes of interest in a weakly supervised context that generalizes robustly to unseen configurations. Localization coupled with a composition classifier significantly outperforms state of the art (SOTA) methods, with an improvement of about 12% on currently available challenging datasets. Further, the modularity enables the use of localized feature extractor to be used with existing OA compositional learning methods to improve their overall performance.

Ekta Prashnani, Orazio Gallo, Joohwan Kim et al.

We tackle the problem of predicting saliency maps for videos of dynamic scenes. We note that the accuracy of the maps reconstructed from the gaze data of a fixed number of observers varies with the frame, as it depends on the content of the scene. This issue is particularly pressing when a limited number of observers are available. In such cases, directly minimizing the discrepancy between the predicted and measured saliency maps, as traditional deep-learning methods do, results in overfitting to the noisy data. We propose a noise-aware training (NAT) paradigm that quantifies and accounts for the uncertainty arising from frame-specific gaze data inaccuracy. We show that NAT is especially advantageous when limited training data is available, with experiments across different models, loss functions, and datasets. We also introduce a video game-based saliency dataset, with rich temporal semantics, and multiple gaze attractors per frame. The dataset and source code are available at https://github.com/NVlabs/NAT-saliency.

Danial Samadi Vahdati, Tai Duc Nguyen, Ekta Prashnani et al.

AI-based talking-head videoconferencing systems reduce bandwidth by sending a compact pose-expression latent and re-synthesizing RGB at the receiver, but this latent can be puppeteered, letting an attacker hijack a victim's likeness in real time. Because every frame is synthetic, deepfake and synthetic video detectors fail outright. To address this security problem, we exploit a key observation: the pose-expression latent inherently contains biometric information of the driving identity. Therefore, we introduce the first biometric leakage defense without ever looking at the reconstructed RGB video: a pose-conditioned, large-margin contrastive encoder that isolates persistent identity cues inside the transmitted latent while cancelling transient pose and expression. A simple cosine test on this disentangled embedding flags illicit identity swaps as the video is rendered. Our experiments on multiple talking-head generation models show that our method consistently outperforms existing puppeteering defenses, operates in real-time, and shows strong generalization to out-of-distribution scenarios.

Riccardo Corvi, Davide Cozzolino, Ekta Prashnani et al.

Synthetic video generation is progressing very rapidly. The latest models can produce very realistic high-resolution videos that are virtually indistinguishable from real ones. Although several video forensic detectors have been recently proposed, they often exhibit poor generalization, which limits their applicability in a real-world scenario. Our key insight to overcome this issue is to guide the detector towards *seeing what really matters*. In fact, a well-designed forensic classifier should focus on identifying intrinsic low-level artifacts introduced by a generative architecture rather than relying on high-level semantic flaws that characterize a specific model. In this work, first, we study different generative architectures, searching and identifying discriminative features that are unbiased, robust to impairments, and shared across models. Then, we introduce a novel forensic-oriented data augmentation strategy based on the wavelet decomposition and replace specific frequency-related bands to drive the model to exploit more relevant forensic cues. Our novel training paradigm improves the generalizability of AI-generated video detectors, without the need for complex algorithms and large datasets that include multiple synthetic generators. To evaluate our approach, we train the detector using data from a single generative model and test it against videos produced by a wide range of other models. Despite its simplicity, our method achieves a significant accuracy improvement over state-of-the-art detectors and obtains excellent results even on very recent generative models, such as NOVA and FLUX.

Ekta Prashnani, Koki Nagano, Shalini De Mello et al.

Modern avatar generators allow anyone to synthesize photorealistic real-time talking avatars, ushering in a new era of avatar-based human communication, such as with immersive AR/VR interactions or videoconferencing with limited bandwidths. Their safe adoption, however, requires a mechanism to verify if the rendered avatar is trustworthy: does it use the appearance of an individual without their consent? We term this task avatar fingerprinting. To tackle it, we first introduce a large-scale dataset of real and synthetic videos of people interacting on a video call, where the synthetic videos are generated using the facial appearance of one person and the expressions of another. We verify the identity driving the expressions in a synthetic video, by learning motion signatures that are independent of the facial appearance shown. Our solution, the first in this space, achieves an average AUC of 0.85. Critical to its practical use, it also generalizes to new generators never seen in training (average AUC of 0.83). The proposed dataset and other resources can be found at: https://research.nvidia.com/labs/nxp/avatar-fingerprinting/.

Ekta Prashnani, Hong Cai, Yasamin Mostofi et al.

The ability to estimate the perceptual error between images is an important problem in computer vision with many applications. Although it has been studied extensively, however, no method currently exists that can robustly predict visual differences like humans. Some previous approaches used hand-coded models, but they fail to model the complexity of the human visual system. Others used machine learning to train models on human-labeled datasets, but creating large, high-quality datasets is difficult because people are unable to assign consistent error labels to distorted images. In this paper, we present a new learning-based method that is the first to predict perceptual image error like human observers. Since it is much easier for people to compare two given images and identify the one more similar to a reference than to assign quality scores to each, we propose a new, large-scale dataset labeled with the probability that humans will prefer one image over another. We then train a deep-learning model using a novel, pairwise-learning framework to predict the preference of one distorted image over the other. Our key observation is that our trained network can then be used separately with only one distorted image and a reference to predict its perceptual error, without ever being trained on explicit human perceptual-error labels. The perceptual error estimated by our new metric, PieAPP, is well-correlated with human opinion. Furthermore, it significantly outperforms existing algorithms, beating the state-of-the-art by almost 3x on our test set in terms of binary error rate, while also generalizing to new kinds of distortions, unlike previous learning-based methods.