AI Paper Atlas

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

GKDT: General Keypoint Detection Transformer

Changsheng Lu, Yuxin Chen, Haokun GUI, Rong Wang, Jie Yang, Harry Yang, Anton van den Hengel, Jiaya Jia

ECCV 2026 · Detection and segmentation › Open-set and few-shot detection · approach: Open-vocabulary object detection · 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
tune the DINOv3 visual backbone (ablation)GKDT without KG transformer · Animal posePCK@0.1 visual: 53.86 -> 79.08Improved
add KG transformer (ablation)GKDT with frozen backbone · Animal posePCK@0.1 visual: 53.86 -> 64.22Improved
add KG transformer (ablation)GKDT with tuned backbone · Animal posePCK@0.1 visual: 79.08 -> 81.11Improved
add KG transformer (ablation)GKDT with frozen backbone · CUBPCK@0.1 visual: 79.55 -> 77.39Hurt
add self-attention in KG transformer (ablation)GKDT · HIntPCK@0.1 visual: 64.74 -> 69.05Improved
add cross-attention in KG transformer (ablation)GKDT · HIntPCK@0.1 visual: 64.74 -> 65.90Improved
use dynamic importance sampling instead of uniform sampling (ablation)OpenKD · HIntPCK@0.1 both: 47.02 -> 50.94Improved
use dynamic importance sampling instead of uniform sampling (ablation)OpenKD · CephalometricPCK@0.1 both: 90.99 -> 99.18Improved
use dynamic importance sampling instead of uniform sampling (ablation)GKDT · HIntPCK@0.1 both: 78.18 -> 81.13Improved
use dynamic importance sampling instead of plain importance sampling (ablation)GKDT · CephalometricPCK@0.1 both: 98.43 -> 99.49Improved
increase generated kernel size from 1x1 to 5x5 (ablation)GKDT · Animal posePCK@0.1 text: 73.27 -> 70.56Hurt
increase generated kernel size from 1x1 to 3x3 (ablation)GKDT · Animal posePCK@0.1 text: 73.27 -> 73.89No 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