AI Paper Atlas

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

Object-level Semantic and Spatial Distillation for Open Vocabulary Detection

Zitong Li, Jinzhuo Wu, Fukang Zhao, Xinyue Wang, Jun Chen-CUG, Zhuo Cheng, Dapeng Luo

ICML 2026 · Detection and segmentation › Open-set and few-shot detection · approach: Open-vocabulary object detection

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 spatial distillation (ablation)DINO-DETR ViT-L/14 · OV-COCOAP50 novel: 38.3 -> 42.3Improved
add semantic distillation (ablation)DINO-DETR ViT-L/14 · OV-COCOAP50 novel: 38.3 -> 45.0Improved
add LQEH (ablation)DINO-DETR ViT-L/14 · OV-COCOAP50 novel: 38.3 -> 41.7Improved
add spatial distillation to semantic distillation (ablation)DINO-DETR ViT-L/14 · OV-COCOAP50 novel: 45.0 -> 47.8Improved
add LQEH to both distillations (ablation)DINO-DETR ViT-L/14 · OV-COCOAP50 novel: 47.8 -> 49.2Improved
swap DETR encoder teacher for DINOv2 as spatial teacher (ablation)OSSD ViT-L/14 · OV-COCOAP50 novel: 49.2 -> 52.4Improved
swap DETR encoder teacher for DINOv2 as spatial teacher (ablation)OSSD ViT-L/14 · OV-LVISmAPr: 40.5 -> 41.8Improved
replace last attention with v-v attention (ablation)CLIP ViT-L/14 · OV-COCO region classificationTop-1 accuracy (all): 40.9 -> 69.4Improved
use v-v instead of k-k last attention (ablation)CLIP ViT-L/14 · OV-COCO region classificationTop-1 accuracy (all): 66.1 -> 69.4Improved
add token prompt on last 5 layers instead of final (ablation)CLIP ViT-L/14 v-v · OV-COCO region classificationTop-1 novel accuracy: 72.2 -> 73.6Improved
add semantic distillation on all layers vs last 3 (ablation)CLIP ViT-L/14 v-v · OV-COCO region classificationTop-1 accuracy (all): 71.0 -> 77.2Improved
add semantic distillation (all layers) versus token prompt last 5 (ablation)CLIP ViT-L/14 v-v · OV-COCO region classificationTop-1 base accuracy: 88.2 -> 76.3Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 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