AI Paper Atlas

Research map / Detection and segmentation / CNN features

MAE-DET: Revisiting Maximum Entropy Principle in Zero-Shot NAS for Efficient Object Detection

Zhenhong Sun, Ming Lin, Xiuyu Sun, Zhiyu Tan, Hao Li, Rong Jin

ICML 2022 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 18 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
Search a backbone with MAE-DET rather than use ResNet-50 at similar computeGFLV2 on COCO · COCO val2017AP_val: 44.7 -> 46.8Improved
Use MAE-DET-M rather than DetNAS-3.8G under matched GFLV2 trainingGFLV2 · COCO val2017AP_val: 46.4 -> 46.8Improved
Use MAE-DET rather than training-based search methodsObject-detection backbone NAS · COCO test-devsearch cost (GPU days): 68 -> 0.6; 26 -> 0.6Improved
Use multi-scale rather than single-scale entropy (ablation)MAE-DET search · COCO with YOLOFAP_val: +0.3%Improved
Use multi-scale rather than single-scale entropy (ablation)MAE-DET search · COCO with FCOSAP_val: 0.6%Improved
Use multi-scale rather than single-scale entropy (ablation)MAE-DET search · ImageNet-1KTOP-1 accuracyNo clear effect
Replace single-scale entropy with Zen-Score (ablation)Zero-shot backbone search · COCO with FCOSAP_val: 38.8 -> 38.1Hurt
Add coarse-to-fine mutation to multi-scale entropy search (ablation)MAE-DET · ImageNet-1KTOP-1 accuracy: 78.9 -> 79.1Improved
Add coarse-to-fine mutation to multi-scale entropy search (ablation)MAE-DET · COCO with YOLOFAP_val: 40.1 -> 40.3Improved
Add coarse-to-fine mutation to multi-scale entropy search (ablation)MAE-DET · COCO with FCOSAP_val: 39.4 -> 40.0Improved
Increase the C5 weight in MSEP, using α=1:1:6 instead of α=1:1:1 (ablation)MAE-DET · COCO val2017, FCOSAP_val: 37.4 -> 39.4Improved
Set MSEP weights to α=1:4:1 rather than α=1:1:6 (ablation)MAE-DET · COCO val2017, FCOSAP_val: 39.4 -> 35.7Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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