SliderEdit++ regularizes slider paths for smoother image edits
Original title: SliderEdit++: Improved Slider Trajectories in Instruction-Based Image Editing
On SlideBench the paper reports an 18.5% higher overall score and about 3x to 20x faster editing.
SliderEdit++ targets abrupt jumps, non-monotonic changes, and interference between jointly controlled sliders in instruction-based image editing. The authors introduce SlideBench, with metrics for monotonic progression, perceptual continuity, and inter-slider disentanglement. The method regularizes intermediate trajectories with velocity, semantic, and spectral controls plus a disentanglement objective, rather than supervising only endpoints. They report state-of-the-art SlideBench scores and faster interactive editing with a few-step backbone.
- SlideBench measures continuity, monotonicity, and slider disentanglement.
- Trajectory regularization covers velocity, semantic space, and spectral control.
- Reported overall score gain is 18.5% versus the strongest baseline.
- A few-step backbone is described as about 3x to 20x faster.
From the original article
1 min read at huggingface.co
Slider-based control aims to give users continuous, fine-grained control over the intensity of individual edits in instruction-based image editing. However, existing approaches fail to deliver reliable continuous controllability: traversing a slider can produce abrupt visual jumps and non-monotonic changes, while jointly controlled sliders can interfere with one another.
By Arman Zarei, Sayan Nag, Bryan Wang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.