Searcharxiv⌕ Search

arXiv subjects

Noemi Roggero

Publications and source records attributed to Noemi Roggero.

2 recordsLinked to original sources

MoSTE: Mixture of Spatio-Temporal Experts for Deepfake Detection in Videos

Generalizable deepfake video detection remains challenging because forgery artifacts vary substantially across manipulation types and often combine diverse spatial and temporal inconsistencies. Existing parameter-efficient adaptation methods typically rely on a single shared representation, rendering them prone to manipulation-specific bias and limiting transfer to unseen forgeries. To address this challenge, we propose Mixture Of Spatio-Temporal Experts (MoSTE), an adaptation framework that enables conditional specialization of forensic representations. Instead of forcing one adapter to model all forgery patterns, MoSTE decomposes adaptation into parallel spatial and temporal expert pools and dynamically routes tokens to the most relevant experts through an unsupervised token-level routing mechanism. This allows experts to naturally specialize in complementary forgery cues without requiring explicit manipulation labels or multi-stage training, while effectively transferring representations from CLIP to the video domain. Extensive experiments across cross-dataset and cross-manipulation benchmarks demonstrate that MoSTE sets a new state of the art in average generalization while preserving strong performance across diverse manipulation families. Additional analysis of expert routing and attention behavior reveals the emergence of expert diversity, suggesting that MoSTE learns complementary spatio-temporal forensic features. Code is available at https://github.com/charbelyah/MoSTE

cs.CV↗

The dipper population of Taurus seen with K2

Dippers are typically low-mass, pre-main-sequence stars that display dips in their light curves. These dips have been attributed to dusty warps that form in the inner part of the disk. Our goal is to derive the properties of dipper stars in Taurus to assess the physical mechanisms that induce dipper light curves. We used the light curves of K2 C4 and C13 to select a dipper sample among 179 members and possible members of the Taurus star-forming region based on the light-curve morphology. We studied the periodicities by combining periodograms with wavelet analysis and derived the stellar parameters from the photometry. We also studied the morphology of the photometric dips. We find a dipper occurrence of ~30% in disk-bearing stars observed with K2 that were identified visually by us. This represents a lower limit to their true occurrence. About half of the dippers are aperiodic, and most of these are dominated by another type of variability. The chosen sample is of late spectral type (K/M), low mass and moderate mass accretion rates and has periods of a few days. We observed a transient dipper over a few rotation cycles and a dipper with a changing period. The structure of the dips can be complex and varies strongly over timescales of down to one stellar rotation. The corotation radii are located at a few stellar radii, and the temperatures at corotation allow dust survival. Many of the systems are seen at moderate to high inclination. We find that the angular extension of the dusty structure producing the dips is correlated with the stellar period. Magnetospheric accretion, which causes an accretion column and its base to occult the star, can explain most of the observed light curves. Although compatible with the model, many of the stellar inclination angles are moderate and do not exclude mechanisms other than the occultation by an inner disk warp to account for dipper light curves.

astro-ph.SR↗