AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Visual Prompting via Image Inpainting

Amir Bar, Yossi Gandelsman, Trevor Darrell, Amir Globerson, Alexei A Efros

NeurIPS 2022 · Detection and segmentation › Open-set and few-shot detection · approach: Object-level representation learning · 50 citations (OpenAlex)

Read the paperCheck 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 changedWhereMeasuredEffect
Pretrain the inpainting model on Computer Vision Figures rather than ImageNet alone (ablation)Visual prompting inpainting models · Downstream image-to-image taskstask-specific metricImproved
Add ImageNet images to Figures pretraining (ablation)MAE-VQGAN · Pascal-5i Foreground SegmentationmIOUImproved
Increase the number of input-output examples in the visual prompt (ablation)MAE-VQGAN (Figures) · Pascal-5i Foreground SegmentationmIOUImproved
Use a vertical rather than horizontal prompt layout (ablation)MAE-VQGAN (Figures) · Pascal-5i Foreground SegmentationmIOU: 27.17 -> 31.57Improved
Use black/white rather than purple/yellow mask colors (ablation)MAE-VQGAN (Figures) · Pascal-5i Foreground Segmentation, Vertical layoutmIOU: 28.47 -> 31.57Improved
Ensemble different prompt layouts (ablation)MAE-VQGAN (Figures) · Synthetic study Colorcolor-aware mIOU: 39.97 -> 44.14Improved
Ensemble different prompt layouts (ablation)MAE-VQGAN (Figures) · Synthetic study Shapecolor-aware mIOU: 46.54 -> 60.42Improved
Ensemble different prompt layouts (ablation)MAE-VQGAN (Figures) · Synthetic study Sizecolor-aware mIOU: 42.01 -> 49.42Improved
Use MAE-VQGAN instead of the strongest reported alternativeVisual prompting · Pascal-5i Foreground SegmentationmIOU: 19.57 -> 27.17Improved
Use MAE-VQGAN instead of the strongest reported alternativeVisual prompting · Pascal VOC 2012 Single Object DetectionmIOU: 5.39 -> 25.00Improved
Use MAE-VQGAN instead of the best reported model for MSEVisual prompting · ImageNet validation ColorizationMSE: 0.43 -> 0.67Hurt
Use MAE-VQGAN instead of the strongest reported alternativeVisual prompting · ImageNet validation ColorizationLPIPS: 0.56 -> 0.40Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy