Research map / Image generation / Editing and style transfer
FED-Bench: A Cross-Granular Benchmark for Disentangled Evaluation of Facial Expression Editing
Fengjian Xue, Xuecheng Wu, Heli Sun, Yunyun Shi, Shi Chen, Liangyu Fu, Jinheng Xie, Dingkang Yang, Hao Wang, Junxiao Xue, Liang He
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 |
|---|---|---|---|
| fine-tune on the 20k+ FED training set | FLUX-Kontext-Dev · FED-Bench (dense instructions) | FED-Score: .243 -> .332 | Improved |
| fine-tune on the 20k+ FED training set | FLUX-Kontext-Dev · FED-Bench (simple instructions) | FED-Score: .120 -> .325 | Improved |
| add Relative Expression Gain dimension to FED-Score (ablation) | FED-Score · Human alignment, overall | agreement with human: 0.7577 -> 0.7700 | Improved |
| add Fidelity dimension (ablation) | FED-Score · Human alignment, overall | agreement with human: 0.7379 -> 0.7700 | Improved |
| add Alignment dimension (ablation) | FED-Score · Human alignment, overall | agreement with human: 0.7279 -> 0.7700 | Improved |
| add MLLM-based model metrics (ablation) | FED-Score · Human alignment, overall | agreement with human: 0.7202 -> 0.7700 | Improved |
| add rule-based metrics (ablation) | FED-Score · Human alignment, overall | agreement with human: 0.7422 -> 0.7700 | Improved |
| use LPIPS ratio to GT instead of raw LPIPS to GT for expression magnitude | REG metric · Human alignment, REG | agreement with human: 0.7045 -> 0.7386 | Improved |
| use FED-Score instead of best single baseline CLIP-I_GT | overall evaluation metric · Human alignment, overall | agreement with human: 0.6850 -> 0.7700 | Improved |
| use ground-truth-referenced variants of metrics | DINO, LPIPS, CLIP-I · Human alignment, overall | agreement with human: 0.5816 -> 0.6216 (DINO) | Improved |
| regroup 7 expressions into 3 polarities (coarse-grained recognition) | Gemini-2.5-Pro · expression recognition | accuracy: 64.93% -> 81.23% | Improved |
| use 5-model voting ensemble instead of 3-model (ablation) | MLLM ensemble · expression recognition (fine) | best accuracy: 66.13 -> 66.67 | 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.
- InstructPix2Pix — InstructPix2Pix: Learning To Follow Image Editing Instructions (CVPR 2023) · better on 2 of 2 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
- LEED: Label-Free Expression Editing via Disentanglement (ECCV 2020 · similarity 0.89)
- Toward Fine-grained Facial Expression Manipulation (ECCV 2020 · similarity 0.88)
- AUEditNet: Dual-Branch Facial Action Unit Intensity Manipulation with Implicit Disentanglement (CVPR 2024 · similarity 0.87)
- DreamSalon: A Staged Diffusion Framework for Preserving Identity-Context in Editable Face Generation (CVPR 2024 · similarity 0.87)
- I2EBench: A Comprehensive Benchmark for Instruction-based Image Editing (NeurIPS 2024 · similarity 0.85)
- Learning Adaptive Spatial Coherent Correlations for Speech-Preserving Facial Expression Manipulation (CVPR 2024 · similarity 0.85)
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.
