AI Paper Atlas

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

Alterbute: Editing Intrinsic Attributes of Objects in Images

Tal Reiss, Daniel Winter, Matan Yaakov Cohen, Alex Rav-Acha, Yael Pritch, Ariel Shamir, Yedid Hoshen

ICML 2026 · Detection and segmentation › Open-set and few-shot detection · approach: Object-level representation learning · 0 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
train for 100K steps instead of 50K (ablation)Alterbute · VLM preference vs baselines (Gemini)preference rate (%): 78.0 -> 86.1Improved
train for 100K steps instead of 50K (ablation)Alterbute · VLM preference vs baselines (Claude)preference rate (%): 76.3 -> 84.9Improved
add grid (1x2 cross-image self-attention) identity conditioning instead of channel-wise concatenation (ablation)Alterbute · Intrinsic attribute editingedit success (qualitative)Improved
define identity by VNE cluster instead of DINOv2 features (ablation)Alterbute · Intrinsic attribute editingedit quality (qualitative)Improved
define identity by VNE cluster instead of instance-retrieval features (ablation)Alterbute · Intrinsic attribute editingedit quality (qualitative)Improved
train with relaxed VNE reference instead of in-place editing (target as its own reference) (ablation)Alterbute · Intrinsic attribute editingedit quality (qualitative)Improved
Alterbute versus FluxKontextAlterbute · VLM preference (Gemini)preference rate (%): 61.2Improved
Alterbute versus Qwen-image-editingAlterbute · VLM preference (Gemini)preference rate (%): 49.8No clear effect
Alterbute versus UltraEdit on conventional DINO metricAlterbute · Conventional protocolDINO: 0.841 -> 0.815Hurt
Alterbute versus Diptych on CLIP-TAlterbute · Conventional protocolCLIP-T: 0.313 -> 0.321Improved

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