DynaDrag: Dynamic Drag-Style Image Editing by Motion Prediction
This work addresses pixel-level image manipulation for users needing precise editing, but it appears incremental as it builds on existing frameworks with specific improvements.
The paper tackles the problem of miss tracking and ambiguous tracking in drag-style image editing by introducing DynaDrag, a method under a predict-and-move framework that iteratively predicts and moves handle points, resulting in improved performance on face and human datasets.
To achieve pixel-level image manipulation, drag-style image editing which edits images using points or trajectories as conditions is attracting widespread attention. Most previous methods follow move-and-track framework, in which miss tracking and ambiguous tracking are unavoidable challenging issues. Other methods under different frameworks suffer from various problems like the huge gap between source image and target edited image as well as unreasonable intermediate point which can lead to low editability. To avoid these problems, we propose DynaDrag, the first dragging method under predict-and-move framework. In DynaDrag, Motion Prediction and Motion Supervision are performed iteratively. In each iteration, Motion Prediction first predicts where the handle points should move, and then Motion Supervision drags them accordingly. We also propose to dynamically adjust the valid handle points to further improve the performance. Experiments on face and human datasets showcase the superiority over previous works.