Visual Examples
Visual Examples
Abstract
Interactive point-based image editing serves as a controllable editor, enabling precise and flexible manipulation of image content. However, most drag-based methods operate primarily on the 2D pixel plane with limited use of 3D cues. As a result, they often produce imprecise and inconsistent edits, particularly in geometry-intensive scenarios such as rotations and perspective transformations. To address these limitations, we propose a novel geometry-guided drag-based image editing method—GeoDrag, which addresses three key challenges: 1) incorporating 3D geometric cues into pixel-level editing, 2) mitigating discontinuities caused by geometry-only guidance, and 3) resolving conflicts arising from multi-point dragging. Built upon a unified displacement field that jointly encodes 3D geometry and 2D spatial priors, GeoDrag enables coherent, high-fidelity, and structure-consistent editing in a single forward pass. In addition, a conflict-free partitioning strategy is introduced to isolate editing regions, effectively preventing interference and ensuring consistency. Extensive experiments across various editing scenarios validate the effectiveness of our method, showing superior precision, structural consistency, and reliable multi-point editability.
Framework
Overall framework of GeoDrag. In drag pipeline, the mask is split into sub-regions, each with a pair of drag points. For each sub-region, the geometry- and plane-aware displacement fields are independently calculated. Subsequently, these fused fields are aggregated without conflict. The final field enables one-step editing via latent relocation and interpolation, with reference guidance to preserve semantics.
Core Insights
- Geometry-Guided Editing: Seamlessly integrates 3D geometric cues with 2D pixel-level editing to enable precise control over complex transformations like rotations and perspective changes.
- Unified Displacement Field: Proposes a unified displacement field that jointly encodes 3D geometry and 2D spatial priors, enabling coherent editing in a single forward pass without requiring multi-step refinement.
- Conflict-Free Multi-Point Editing: Introduces a conflict-free partitioning strategy that isolates editing regions, effectively preventing interference and ensuring consistency when handling multiple simultaneous edits.
- Structure-Preserving Consistency: Achieves high-fidelity, structure-consistent editing that maintains semantic coherence and prevents artifacts caused by latent manipulation.
Experiments
🏆 Main Results on DragBench
| Method | MD ↓ | DAI₁ ↓ | DAI₁₀ ↓ | DAI₂₀ ↓ | IF ↑ | Preparation | Time (s) | Mem (GB) |
|---|---|---|---|---|---|---|---|---|
| DragDiffusion | 34.57 | 0.181 | 0.170 | 0.160 | 0.871 | ~1 min (LoRA) | 22.46 | 18.63 |
| FreeDrag | 30.80 | 0.183 | 0.166 | 0.151 | 0.845 | ~1 min (LoRA) | 42.90 | 18.90 |
| CLIPDrag | 34.62 | 0.195 | 0.174 | 0.158 | 0.891 | ~1 min (LoRA) | 38.21 | 22.72 |
| AdaptiveDrag | 32.38 | 0.180 | 0.154 | 0.146 | 0.830 | ~1 min (LoRA) | 46.30 | 7.71 |
| DragNoise | 33.84 | 0.179 | 0.169 | 0.158 | 0.861 | ~1 min (LoRA) | 21.12 | 18.36 |
| FastDrag | 32.10 | 0.131 | 0.123 | 0.115 | 0.850 | – | 3.23 | 5.85 |
| GeoDrag (Ours) | 29.24 | 0.128 | 0.120 | 0.111 | 0.847 | – | 3.95 | 5.44 |
Quantitative results on DragBench. Lower MD and DAI indicate higher editing precision, and higher IF reflects greater similarity between original and edited images. Time is the average editing time per point (in seconds), and Mem is the peak GPU memory in GB. GeoDrag achieves the best precision with competitive efficiency.
📊 Qualitative Comparisons
Qualitative comparisons with the state-of-the-art interactive point-based methods. Red points mark handles, and blue points mark targets.
BibTeX
@inproceedings{
pu2026dragging,
title={Dragging with Geometry: From Pixels to Geometry-Guided Image Editing},
author={Xinyu Pu and Hongsong Wang and Jie Gui and Pan Zhou},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=MBiMt3wp8M}
}