AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

UniKD: Universal Knowledge Distillation for Mimicking Homogeneous or Heterogeneous Object Detectors

Shanshan Lao, Guanglu Song, Boxiao Liu, Yu Liu, Yujiu Yang

ICCV 2023 · Label-efficient and robust learning › Knowledge distillation · approach: Knowledge distillation for detectors · 11 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 both content and position queries instead of one type (ablation)UniKD · COCO RetinaNet R50 -> R18mAP: 34.9 / 34.5 -> 35.3Improved
combine UniKD with FitNet feature mimicking (ablation)UniKD · COCO RetinaNet R50 -> R18mAP: 35.3 -> 34.9Hurt
increase queries from 200 to 300 (ablation)UniKD · COCO RetinaNet R18 studentmAP: 35.3 -> 35.2No clear effect
use 3 cross-attention layers instead of 1 in fe (ablation)UniKD · COCO RetinaNet R18 studentmAP: 35.3 -> 34.9Hurt
pretrain AKE for 15 instead of 5 epochs (ablation)UniKD · COCO RetinaNet R18 studentmAP: 35.2 -> 35.3No clear effect
add 2 self-attention adaptor layers (ablation)UniKD with Deformable DETR teacher · COCO RetinaNet R18 studentmAP: 31.9 -> 33.8Improved
add self-attention adaptor layers to FitNet (ablation)FitNet with Deformable DETR teacher · COCO RetinaNet R18 studentmAP: 31.8 -> 31.3Hurt
replace feature mimicking with UniKD for heterogeneous teacherRetinaNet R18 student · COCO heterogeneous pairsmAP: FGD 31.3 -> UniKD 34.8Improved
use FitNet with Deformable DETR teacherRetinaNet R18 student · COCOmAPHurt
use a stronger teacher with different architectureUniKD · COCO Table 5mAPNo clear effect

Later papers that compared with this method

1 later accepted paper reports UNI-KD as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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