Research Demo · SRM Institute of Science and Technology

CrossFakeNet

Multimodal fake news detection — select input type to route through the appropriate AI model pipeline

// select input modality
Pipeline: RoBERTa-Large → Sentiment → Credibility Scorer
Pipeline: Whisper ASR → Wav2Vec2 prosody → RoBERTa-Large
🎙️
Click to upload

Supports MP3, WAV, M4A, OGG — max 50MB

Pipeline: ViT-L/16 → CLIP alignment → GradCAM manipulation detection
🖼️
Click to upload

Supports JPG, PNG, WEBP — max 10MB

Pipeline: ViT-L/16 + RoBERTa → CMAF Cross-Attention → Social GAT
💬
Click to upload

Supports JPG, PNG, WEBP — max 10MB

Pipeline: ViT frames + Whisper + Wav2Vec2 → BiLSTM-GCN → CMAF fusion
🎬
Click to upload

Supports MP4, AVI, MOV — max 100MB

// models in this system

🔤
RoBERTa-Large
Text encoding (text, audio, video)
👁️
ViT-L/16
Visual encoding (image, video)
🎙️
Whisper ASR
Audio transcription (audio, video)
🔊
Wav2Vec2
Prosody features (audio, video)
🔗
CLIP proxy
Image–text alignment (image)
🕸️
Graph ATN (GAT)
Social context (image+comments, video)
🔀
CMAF module
Cross-modal fusion (image+comments, video)
⏱️
BiLSTM-GCN
Temporal + spatial encoding (video)