Research map / Video / Video segmentation and tracking
DREAM: A Benchmark Study for Deepfake PhotoRealism AssessMent
Bo Peng, Zichuan Wang, Yu Sheng, Xiaochuan Jin, Wei Wang, Jing Dong
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 |
|---|---|---|---|
| add description-aligned CLIP training with textual branch to the OPDAI visual backbone and loss | OPDAI backbone · DREAM | avr: 0.782 -> 0.827 | Improved |
| pretrain on deepfake detection data instead of ImageNet (ablation) | OPDAI · DREAM | avr: 0.750 -> 0.782 | Improved |
| pretrain on deepfake detection data instead of ImageNet (ablation) | UNILJ · DREAM | avr: 0.611 -> 0.702 | Improved |
| replace MAE loss with RMSE loss (ablation) | OPDAI · DREAM | avr: 0.631 -> 0.725 | Improved |
| replace RMSE loss with KL loss (ablation) | OPDAI · DREAM | avr: 0.725 -> 0.810 | Improved |
| combine NinN with KL instead of KL alone (ablation) | OPDAI · DREAM | avr: 0.810 -> 0.782 | Hurt |
| add PLCC loss to RMSE (ablation) | OPDAI · DREAM | avr: 0.725 -> 0.788 | Improved |
| add rank loss to RMSE (ablation) | OPDAI · DREAM | avr: 0.725 -> 0.726 | No clear effect |
| use InternVL2.5-8B backbone instead of mPLUG-Owl2 (ablation) | Q-Align · DREAM | avr: 0.203 -> 0.644 | Improved |
| use InternVL2.5-8B backbone instead of mPLUG-Owl2 (ablation) | DeQA-Score · DREAM | avr: 0.213 -> 0.692 | Improved |
| use deepfake-detection features (DFGC1st) instead of ResNet50 features | feature-based regression · DREAM | avr: 0.446 -> 0.682 | Improved |
| use top-11 retrieved descriptions instead of random-1 for explanation (ablation) | DA-CLIP explanation · DREAM explanation error | RMSE (lower better): 0.099 -> 0.063 | Improved |
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.
- Fast-VQA — FAST-VQA: Efficient End-to-End Video Quality Assessment with Fragment Sampling (ECCV 2022) · better on 1 of 1 shared result
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
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · similarity 0.89)
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes (NeurIPS 2025 · similarity 0.88)
- TriDF: Evaluating Perception, Detection, and Hallucination for Interpretable DeepFake Detection (CVPR 2026 · similarity 0.88)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.87)
- Through the Lens: Benchmarking Deepfake Detectors Against Moiré-Induced Distortions (NeurIPS 2025 · similarity 0.87)
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.87)
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.
