Research map / Vision-language and multimodal / Cross-modal learning
Intra-modal and Cross-modal Synchronization for Audio-visual Deepfake Detection and Temporal Localization
Ashutosh Anshul, Shreyas Gopal, Deepu Rajan, Eng Siong Chng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Gaussian-target MSE loss instead of InfoNCE (ablation) | Decoder, VoxCeleb2 pretraining · FakeAVCeleb | AUC: 80.1 -> 88.0 | Improved |
| use Gaussian-target MSE loss instead of InfoNCE (ablation) | Decoder, VoxCeleb2 pretraining · FakeAVCeleb | AP: 99.4 -> 99.7 | Improved |
| add intra-modal synchronization (AA, VV) pretraining (ablation) | Decoder, VoxCeleb2 pretraining · FakeAVCeleb | AUC: 79.7 -> 88.0 | Improved |
| add intra-modal synchronization (AA, VV) pretraining (ablation) | Decoder, VoxCeleb2 pretraining · FakeAVCeleb | AP: 99.4 -> 99.7 | Improved |
| use unimodal embeddings instead of intra-modal sync features in stage two (ablation) | classification stage · FakeAVCeleb | AP/AUC: +0.1% AP, +3.8% AUC | Improved |
| use sparse encoder instead of decoder attention variant (ablation) | synchronization model · FakeAVCeleb and KoDF | AP/AUC: 8 hrs -> 2 hrs per epoch | No clear effect |
| use sparse encoder instead of decoder attention variant (ablation) | Ours (VoxCeleb2) · FakeAVCeleb RVFA | AP: 98.6 -> 96.5 | Hurt |
| pretrain on VoxCeleb2 instead of LRS2 (decoder) (ablation) | Decoder variant · KoDF | AP: 56.7 -> 98.9 | Improved |
| pretrain on VoxCeleb2 instead of LRS2 (sparse encoder) (ablation) | Sparse Encoder variant · KoDF | AP: 98.6 -> 94.4 | Hurt |
| use Decoder VoxCeleb2 model instead of AVFF | deepfake classification · FakeAVCeleb RVFA | AP: 93.3 -> 98.6 | Improved |
| use Decoder VoxCeleb2 model instead of UMMAFormer | temporal localization · LAV-DF | AP@0.95: 37.61 -> 5.72 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AVFF — AVFF: Audio-Visual Feature Fusion for Video Deepfake Detection (CVPR 2024) · better on 4 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Multi-modal Deepfake Detection via Multi-task Audio-Visual Prompt Learning (AAAI 2025 · similarity 0.91)
- AVFF: Audio-Visual Feature Fusion for Video Deepfake Detection (CVPR 2024 · similarity 0.90)
- Circumventing Shortcuts in Audio-visual Deepfake Detection Datasets with Unsupervised Learning (CVPR 2025 · similarity 0.87)
- Dense Audio-Visual Event Localization Under Cross-Modal Consistency and Multi-Temporal Granularity Collaboration (AAAI 2025 · similarity 0.87)
- FAVE: A Structured Benchmark for Fine-Grained Audio-Visual Temporal Evaluation in Multimodal LLMs (CVPR 2026 · similarity 0.86)
- CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
