Research map / Image generation / Editing and style transfer
SpaceEdit: Learning a Unified Editing Space for Open-Domain Image Color Editing
Jing Shi, Ning Xu, Haitian Zheng, Alex Smith, Jiebo Luo, Chenliang Xu
Check 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 the unified multimodal editing generator instead of BiCycleGAN or DivCo | open-domain image editing on Discover · Discover test | FID and LPIPS: FID: 9.9586 -> 5.1755; LPIPS: 0.1705 -> 0.1945 (DivCo); FID: 12.2837 -> 5.1755; LPIPS: 0.0857 -> 0.1945 (BiCycleGAN) | Improved |
| use the pretrained editing generator and predict an editing latent from image and language | supervised LGIE · MA5k-Req test | L1, SSIM, and FID: L1: 0.0784 -> 0.0731; SSIM: 0.8459 -> 0.8721; FID: 6.7571 -> 5.9791 | Improved |
| use cosine distance instead of euclidean distance for clustering latent editing codes | W-space editing-style clustering · Discover dataset | Purity: 11.30 -> 12.76 | Improved |
| cluster editing styles in W space rather than Lightroom operation space | editing-style clustering · Discover dataset | Purity: 4.25 -> 12.76 | Improved |
| remove deeper encoder and decoder layers (ablation) | multimodal editing generator · Discover test | FID: 5.1755 -> 6.0958 | Hurt |
| remove deeper encoder and decoder layers (ablation) | multimodal editing generator · Discover test | LPIPS: 0.1945 -> 0.1581 | Hurt |
| co-modulate the decoder with input-image features (ablation) | multimodal editing generator · Discover test | FID: 5.1755 -> 5.6355 | Hurt |
| co-modulate the decoder with input-image features (ablation) | multimodal editing generator · Discover test | LPIPS: 0.1945 -> 0.1479 | Hurt |
| remove visual input when predicting the editing latent (ablation) | supervised LGIE · MA5k-Req test | L1, SSIM, and FID: L1: 0.0731 -> 0.0795; SSIM: 0.8721 -> 0.8596; FID: 5.9791 -> 6.9757 | Hurt |
| remove visual input when predicting the editing latent (ablation) | supervised LGIE · MA5k-Req test | image variance (σ×10²): 0.6809 -> 0.6281 | 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.
- DivCo — DivCo: Diverse Conditional Image Synthesis via Contrastive Generative Adversarial Network (CVPR 2021) · 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
- Open-Edit: Open-Domain Image Manipulation with Open-Vocabulary Instructions (ECCV 2020 · similarity 0.89)
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.86)
- FireEdit: Fine-grained Instruction-based Image Editing via Region-aware Vision Language Model (CVPR 2025 · similarity 0.85)
- UltraEdit: Instruction-based Fine-Grained Image Editing at Scale (NeurIPS 2024 · similarity 0.85)
- AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea (CVPR 2025 · similarity 0.85)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · 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.
