Facialabuse-gaia-3 — Better

That being said, I can propose a general outline for a paper on facial abuse in the context of Gaia-3:

| Metric | GAIA‑3 (paper) | GAIA‑2 (baseline) | State‑of‑the‑art (e.g., DeepFakeDetect‑V2) | |--------|----------------|-------------------|-------------------------------------------| | | 0.96 (overall) | 0.92 | 0.95 | | Video‑level AUROC | 0.94 (30 s clips) | 0.89 | 0.93 | | Per‑category F1 (average) | 0.88 | 0.78 | 0.85 | | Inference latency (GPU RTX 3080) | 45 ms / image, 210 ms / 10‑frame clip | 38 ms / image, 180 ms / clip | 38 ms / image, 190 ms / clip | | On‑device (Apple A14) | 210 ms / image (CPU) | 170 ms / image | N/A (no official on‑device support) | Facialabuse-gaia-3

| Strengths | Limitations | |-----------|-------------| | • State‑of‑the‑art detection performance (AUROC ≥ 0.94).• Multimodal (image + short video) support.• Prompt‑based zero‑shot adaptability.• Open‑source, well‑documented code and model card.• On‑device inference option for privacy. | • Large model size; heavy compute for real‑time video.• Temporal window limited to ≤ 30 s.• Slight bias in certain sub‑categories (e.g., forced distortion).• Explanations sometimes generic, not always actionable.• No built‑in adversarial robustness against targeted evasion. | That being said, I can propose a general