图像编辑论文趋势 2023–2026
指令编辑
统计范围：NeurIPS、ICLR、ICML、CVPR 四个会议，2023–2026 年，共 692 篇被归入该主题的论文。
主要结论
- 图像编辑论文从 2023 年的 112 篇增长到 2026 年的 238 篇，三年间扩大到约 2.1 倍。
- CVPR 是发图像编辑论文最多的会议，2023–2026 年共 364 篇，占四个会议总量的 53%。
- 交叉方向里涨得最快的是应用 · computer vision：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 15 篇。
- 在「多模态生成」的 5 个主题里，它的 2026 年论文数排第 3；同组增长最快的叙事生成是 15 倍，图像编辑是 2.1 倍。
各会议论文数
按年份和会议统计
| 年份 | NeurIPS | ICLR | ICML | CVPR | 合计 |
|---|---|---|---|---|---|
| 2023 | 25 | 15 | 8 | 64 | 112 |
| 2024 | 33 | 24 | 13 | 90 | 160 |
| 2025 | 45 | 29 | 11 | 97 | 182 |
| 2026 | 52 | 40 | 33 | 113 | 238 |
代表论文
按年份和会议均匀挑选，方便快速看到每一届的样貌
- NeurIPS 2026Accelerated Image Editing via Consistency-Aware Source Token Pruning
- ICML 2026Klein Hyperbolic Metric Learning
- ICLR 2026DiffVax: Optimization-Free Image Immunization Against Diffusion-Based Editing
- CVPR 2026AnchorFlow Training-Free 3D Editing via Latent Anchor-Aligned Flows
- NeurIPS 2025A$^3$E: Towards Compositional Model Editing
- ICML 2025DEALing with Image Reconstruction: Deep Attentive Least Squares
- ICLR 2025HQ-Edit: A High-Quality Dataset for Instruction-based Image Editing
- CVPR 2025AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea
- NeurIPS 2024Reproducibility study of “LICO: Explainable Models with Language-Image Consistency"
- ICML 2024Counterfactual Image Editing
- ICLR 2024Guiding Instruction-based Image Editing via Multimodal Large Language Models
- CVPR 2024DiffEditor: Boosting Accuracy and Flexibility on Diffusion-based Image Editing
- NeurIPS 2023ImageBrush: Learning Visual In-Context Instructions for Exemplar-Based Image Manipulation
- ICML 2023Random Shuffle Transformer for Image Restoration
- ICLR 2023Editing models with task arithmetic
- CVPR 20233D GAN Inversion With Facial Symmetry Prior
- NeurIPS 2026AnyBokeh: Physics-Guided Any-to-Any Bokeh Editing with Optical Fingerprint Transfer
- ICML 2026DLEBench: Evaluating Small-scale Object Editing Ability for Instruction-based Image Editing Model
- ICLR 2026Energy-Regularized Sequential Model Editing on Hyperspheres
- CVPR 2026ChordEdit One-Step Low-Energy Transport for Image Editing
全部图像编辑论文
共 238 篇，第 5 / 5 页
- NeurIPS 2026EditJudge-Bench: Auditing VLM Image-Edit Judges with Synthetic Ground Truth
- NeurIPS 2026FlowSteer: Towards Agents Designing Agentic Workflows via Reinforced Progressive Canvas Editing
- NeurIPS 2026FlowTrack: Controlling Edit-Signal Execution in Inversion-Free Flow Video Editing
- NeurIPS 2026FreqCa: Accelerating Image generation and editing via Frequency-Aware Caching
- NeurIPS 2026From Structural Feedback to Prompt Policies: Learning Faithful Text-to-Image Prompt Editors
- NeurIPS 2026GenScale: A Benchmark for Relative Object Scale in Image Generation and Editing
- NeurIPS 2026GPT-Image-Edit-1M: An Auditable Million-Scale Dataset for Instruction-Guided Image Editing
- NeurIPS 2026HandEdit: A Unified Benchmark for Egocentric Human-to-Robot Dexterous Hand Image Editing
- NeurIPS 2026IdealCache: Rethinking Cache Scheduling in Diffusion Transformers via Ideal Trajectories
- NeurIPS 2026Inline Critic Steers Image Editing
- NeurIPS 2026Internalize External Competence for Visual Instruction Editing
- NeurIPS 2026Lang-SVG: Hierarchical Image Vectorization with Language Priors
- NeurIPS 2026Language Models Can Coarsely Modulate Entropy Under Instruction
- NeurIPS 2026MetaKE: Meta-Learning for Knowledge Editing Toward a Better Accuracy-Editability Trade-off
- NeurIPS 2026Mission Impossible: Diagnosing and Fixing Non-Operative Instruction Following in Image Editing
- NeurIPS 2026MLLM-Edit: Benchmarking Image Forgery Detection and Localization under MLLM-based Editing
- NeurIPS 2026MorphGen: Controllable Cell-Image Generation with Biological Representation Alignment
- NeurIPS 2026OTEdit: Entropic Transport Corrected Trajectory for Inversion-Free Flow-Based Image Editing
- NeurIPS 2026PAVE: Prefill-Conditioned Activation Editing for Hallucination Mitigation in LVLMs
- NeurIPS 2026Probing Visual Planning in Image Editing Models
- NeurIPS 2026RADIUM: RadioActive Decay of Image-Underlaid Marks
- NeurIPS 2026Real In, Real Out: What If We Only Use Real Data for Scene Text Editing?
- NeurIPS 2026Residual Paving: Diagnosing the Routing Bottleneck in Selective Refusal Editing
- NeurIPS 2026Rethinking Contrastive Targets in Radiology Language-Image Pretraining
- NeurIPS 2026RiSE: Residual Subspace Expert for Generalizable Text-Centric Image Forgery Localization
- NeurIPS 2026SCULPT: Advancing Masked Discrete Diffusion for High-Resolution Image Synthesis.
- NeurIPS 2026SIGMA: Semantic-Difference Instruction-Grounding Mask Annotator for Text-Driven Image Manipulation Localization
- NeurIPS 2026SpatialFlow-GRPO: Where Spatial Credit Drives Image Editing
- NeurIPS 2026Steer2Edit: From Activation Steering to Component-Level Editing
- NeurIPS 2026StitchEdit: Stitching Depth Priors into Image Editors via Per-Layer Gradient Probing
- NeurIPS 2026Text-Vision Co-Instructed Image Editing
- NeurIPS 2026The Fractured Interlingua: Geometric Bottlenecks in Cross-Lingual Knowledge Editing
- NeurIPS 2026Thinking in Boxes: 3D Editing in Real Images Made Easy
- NeurIPS 2026TokenRepel: Generating Diverse Image Sets Without Sacrificing Quality
- NeurIPS 2026Tool-IQA: Augmenting Image Quality Assessment with Simple Tools
- NeurIPS 2026Towards Realistic Conversational Multimodal Instruction Following
- NeurIPS 2026VisEditBench: A Benchmark for Vector-Format Diagram Editing with Visual Instructions
- NeurIPS 2026WordEval: Evaluating Word-Native Operation Fidelity in Document Editing
在 Claude、Cursor 里直接问这个趋势
接上 Paperlist MCP 后，写综述、挑方向或准备投稿时，不用离开对话框就能拿到这页的数字：每年各会议的论文数、增长倍数、涨得最快的子方向和代表论文，AI 会按你的问题整理。
可以直接复制这句话问 AI
用 paperlist 查图像编辑在 CVPR 2023–2026 的论文趋势，哪一年涨得最快？
AI 会调用的工具：get_topic_stats {"topic": "image-editing"}
常见问题
- 2023–2026 年四大会议一共有多少篇图像编辑论文？
- 共 692 篇，按年份分别是 2023 年 112 篇、2024 年 160 篇、2025 年 182 篇、2026 年 238 篇。
- NeurIPS 2026 有多少篇图像编辑论文？
- NeurIPS 2026 有 52 篇图像编辑论文，上一届是 45 篇。
- ICLR 2026 有多少篇图像编辑论文？
- ICLR 2026 有 40 篇图像编辑论文，上一届是 29 篇。
- ICML 2026 有多少篇图像编辑论文？
- ICML 2026 有 33 篇图像编辑论文，上一届是 11 篇。
- CVPR 2026 有多少篇图像编辑论文？
- CVPR 2026 有 113 篇图像编辑论文，上一届是 97 篇。
- 哪个会议的图像编辑论文最多？
- CVPR，2023–2026 年共 364 篇。