Probing anomalous Zγγγ couplings at a future muon collider
Paper • 2403.07689 • Published
Datasets:
IMAGE dict | VIDEO dict | AUDIO dict | HUMAN_BIOMETRIC dict |
|---|---|---|---|
{
"dima806_deepfake": {
"name": "Deepfake vs Real Image Detection",
"type": "huggingface",
"hf_path": "dima806/deepfake_vs_real_image_detection",
"description": "Pre-annotated 140k real & fake face images from StyleGAN/FFHQ."
},
"ntire_robust_detection": {
"name": "NTIRE Robust AI-Generated De... | {
"faceforensics_pp": {
"name": "FaceForensics++ (FF++)",
"type": "external/github",
"repo": "https://github.com/ondyari/FaceForensics",
"description": "1,000 real videos & 5,000 fake videos (Deepfakes, Face2Face, FaceSwap, NeuralTextures)."
},
"celeb_df_v2": {
"name": "Celeb-DF (v2)",
"ty... | {
"deepspeak_audio": {
"name": "DeepSpeak Voice Deepfakes",
"type": "huggingface",
"hf_path": "faridlab/deepspeak_v1",
"description": "Voice clone & audio deepfake speech clips."
},
"fakeavceleb_audio": {
"name": "FakeAVCeleb (Audio-Visual Deepfakes)",
"type": "github",
"repo": "https:... | {
"multifakeverse": {
"name": "MultiFakeVerse (Person-Centric VLM Manipulation)",
"type": "external/arxiv",
"scale": ">845,000 images",
"paper": "https://arxiv.org/abs/2403.07689",
"description": "Large-scale person-centric benchmark using Vision-Language Models (VLMs) to create subtle context, ac... |
Dual-Stream DeepFake Foundational Baseline Connectors
This repository provides standardized data connectors, download manifests, and partition splits for the 8 foundational baseline datasets used in the Dual-Stream Deepfake Detection Framework.
📊 Dual-Stream Model Allocation
🖼️ Model 1: General Vision & Signal Model (>578,000 samples)
- NTIRE-RobustAIGenDetection (~120,000 samples): Multi-generator synthetic artifacts (MSU 2024).
- CIFAKE (120,000 samples): 60,000 real CIFAR-10 vs 60,000 Stable Diffusion.
- DF40 (40,000 samples): 40 advanced deepfake generation algorithms.
- DFFD (298,000 samples): Diverse Fake Face Dataset (58k real / 240k GAN fakes). Augmented with Threads Hard Negatives ($H(P) \ge 0.70$).
🧬 Model 2: Human Biometric Model (>1,105,000 samples)
- MultiFakeVerse (>845,000 samples): Person-centric VLM manipulation (arXiv:2403.07689).
- HiDF (~70,000 samples): Human-Indistinguishable Deepfakes (62k img, 8k vid).
- 140k Real & Fake Faces (140,000 samples): 70k Flickr real vs 70k StyleGAN.
- DeepDetect-2025/2026 (50,000 samples): Adversarial Flux/Midjourney v6 vs Unsplash. Augmented with Threads Vietnamese 5-Taxonomy In-The-Wild Media.
- Downloads last month
- 45