Joint Spatiotemporal Adversarial Attacks on Video Transformer Models Through XAI-guided Perturbation
Bibliographic record
Abstract
The widespread deployment of video transformer models in action recognition systems necessitates a comprehensive understanding of their vulnerability to adversarial attacks. Unlike traditional CNN-based video models, transformers process spatiotemporal dependencies through self-attention mechanisms, creating a different vulnerability profile to adversarial attacks. This study presents an investigation of adversarial robustness in video transformers. We develop a novel joint spatiotemporal attack method that precisely targets the attention mechanisms of video transformers. By simultaneously perturbing both spatial and temporal features, our method achieves a 76.30% in ASR on the Kinetics-400 dataset, outperforming frame-wise attacks and state-of-the-art query-based attacks. To interpret the mechanisms underlying these attacks, we introduce quantitative metrics based on Explainable AI (XAI) analysis. Spatial analysis reveals systematic disruption of attention patterns, with adversarial examples showing median SSIM scores of 0.353. Temporal correlation analysis also demonstrates severe degradation in attention coherence across frame sequences. Through experiments comparing previous attack methods, including common corruptions benchmark, frame-wise attacks, sparse attacks, and recent V-BAD attacks, we demonstrate that our proposed method is more effective in transformer-based video models. This study further examines the adversarial training strategy against the selected attacks. To promote reproducibility and facilitate future research, we provide our methods and analysis tools through a public GitHub repository. These findings underscore the effectiveness of jointly considering spatial and temporal dimensions when developing adversarial attack strategies and defense mechanisms for video AI models.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".