Research map / Video / Video generation
FiVE-Bench: A Fine-grained Video Editing Benchmark for Evaluating Emerging Diffusion and Rectified Flow Models
Minghan Li, Chenxi Xie, Yichen Wu, Lei Zhang, Mengyu Wang
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 |
|---|---|---|---|
| replace diffusion-based editing with RF-based Wan-Edit (inversion-free FlowEdit) | video editing · FiVE benchmark | time per frame (s): 27.12 -> 3.07 | Improved |
| use Wan-Edit instead of VideoGrain for background preservation | Wan-Edit · FiVE benchmark | LPIPS x10^3: 185.21 -> 94.61 | Improved |
| use Wan-Edit instead of VideoGrain for PSNR/MSE | Wan-Edit · FiVE benchmark | PSNR: 27.05 -> 25.57 | Hurt |
| CLIP score as editing-quality metric versus VLM-based FiVE-Acc | evaluation metrics · FiVE benchmark | metric discriminability: TokenFlow 27.43 vs DMT 48.42 FiVE-Acc despite close CLIP scores | Improved |
| RF-based Wan-Edit versus feature-propagation methods on non-rigid edits | Wan-Edit · FiVE Edit2 (non-rigid) | FiVE-Acc (%): 52.02 -> 52.02 (Wan-Edit stable) while DMT 67.86 -> 53.72 | Improved |
| use Pyramid-Edit's multi-resolution autoregressive architecture | Pyramid-Edit · FiVE benchmark | qualitative edit quality | Hurt |
| Wan-Edit versus Pyramid-Edit on object addition | RF-based editing · FiVE Edit5 (object addition) | FiVE-Acc (%): DMT 61% vs Pyramid-Edit 83%, Wan-Edit 72% | Improved |
| object removal editing with current methods | all methods · FiVE Edit6 (object removal) | FiVE-Acc: below 20% | Hurt |
| Qwen-2.5-VL evaluator replacing human evaluator (ablation) | FiVE-Acc · FiVE human validation | FiVE-Acc (%): 47.34 -> 46.97 | No clear effect |
| propagating tokens via feature clustering (TokenFlow, VidToMe) | TokenFlow, VidToMe · FiVE benchmark | FiVE-Acc (%): 27.43 and 26.77 vs 48.42 for DMT | 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.
- VideoGrain — VideoGrain: Modulating Space-Time Attention for Multi-Grained Video Editing (ICLR 2025) · better on 3 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
- EditBoard: Towards a Comprehensive Evaluation Benchmark for Text-Based Video Editing Models (AAAI 2025 · similarity 0.90)
- VE-Bench: Subjective-Aligned Benchmark Suite for Text-Driven Video Editing Quality Assessment (AAAI 2025 · similarity 0.89)
- VEU-Bench: Towards Comprehensive Understanding of Video Editing (CVPR 2025 · similarity 0.89)
- FADE: Frequency-Aware Diffusion Model Factorization for Video Editing (CVPR 2025 · similarity 0.88)
- ViTeX-Bench: Benchmarking High-Fidelity Video Scene Text Editing (NeurIPS 2026 · similarity 0.88)
- TokenFlow: Consistent Diffusion Features for Consistent Video Editing (ICLR 2024 · similarity 0.88)
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.
