AI Paper Atlas

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

T-Rex-Omni: Integrating Negative Visual Prompt in Generic Object Detection

Jiazhou Zhou, Qing Jiang, Kanghao Chen, Lutao Jiang, Yuanhuiyi Lyu, Ying-Cong Chen, Lei Zhang

AAAI 2026 · Detection and segmentation › Open-set and few-shot detection · approach: Open-vocabulary object detection · 2 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
add training-free NNC module (ablation)T-Rex2 (Swin-T) · COCO val (zero-shot)AP: 38.8 -> 41.8Improved
add training-free NNC module (ablation)T-Rex2 (Swin-T) · LVIS minival (zero-shot)AP: 37.4 -> 40.6Improved
fine-tune with NNC (ablation)T-Rex-Omni · COCO val (zero-shot)AP: 41.8 -> 42.9Improved
add NNH hinge loss (ablation)T-Rex-Omni · COCO val (zero-shot)AP: 42.9 -> 43.6Improved
add NNH hinge loss (ablation)T-Rex-Omni · LVIS minival (zero-shot)AP: 41.4 -> 43.0Improved
stochastic mode switching B~Bernoulli(0.5) instead of fixed B=1 in training (ablation)T-Rex-Omni · COCO val (zero-shot)AP: 40.8 -> 42.0 (eval B=0); 42.4 -> 43.6 (eval B=1)Improved
set beta=0.3 vs no suppression (beta=0.0) (ablation)NNC module · COCO val (zero-shot)AP: 39.7 -> 42.8Improved
over-suppress negatives (beta >= 0.5) (ablation)NNC module · COCO val (zero-shot)APHurt
margin eta 0.1-0.3 vs no margin in NNH loss (ablation)NNH loss · COCO val (zero-shot)APImproved
excessive NNH margin (eta >= 0.5) (ablation)NNH loss · COCO val (zero-shot)APHurt
use three negative prompts instead of one (ablation)T-Rex-Omni · COCO val (zero-shot)AP: +0.6 AP; 43.6Improved
increase negative prompts from three to five (ablation)T-Rex-Omni · COCO val (zero-shot)AP: 43.6 -> 43.5No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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