AI Paper Atlas

Research map / Detection and segmentation / CNN features

NAS-FCOS: Efficient Search for Object Detection Architectures

Ning Wang, Yang Gao, Hao Chen, Peng Wang, Zhi Tian, Chunhua Shen, Yanning Zhang

IJCV 2021 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 30 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 both FPN and head (progressive) instead of head only (ablation)NAS-FCOS (VOC) R-50 · COCO test-devAP: 38.7 -> 39.8Improved
search both FPN and head instead of FPN only (ablation)NAS-FCOS (VOC) R-50 · COCO test-devAP: 38.9 -> 39.8Improved
use searched space instead of deformable convs in FPN and head (ablation)NAS-FCOS (VOC) R-50 · COCO test-devAP: 38.7 -> 39.8Improved
use searched FPN instead of deformable-conv FPN (ablation)NAS-FCOS (VOC) R-50 · COCO test-devAP: 38.4 -> 38.9Improved
replace standard convs in FCOS with deformable convs (ablation)FCOS R-50 · COCO test-devAP: 37.4 -> 38.7Improved
use negative detection loss as RL reward instead of AP (ablation)NAS controller · VOC proxy taskAPImproved
use COCO-subset proxy instead of VOC proxyNAS-FCOS @256 R-50 · COCO test-devAP: 39.8 -> 40.3Improved
use COCO-subset proxy instead of VOC proxyNAS-FCOS @256 MobileNetV2 · COCO test-devAP: 34.7 -> 36.2Improved
replace FPN-FCOS decoder by searched decoderNAS-FCOS (COCO) R-50 · COCO test-devAP: 37.4 -> 40.3Improved
transfer searched structure to RetinaNetNAS-RetinaNet R-50 · COCOAP: 36.1 -> 39.5Improved
use 128-channel searched decoder instead of FCOS decoderNAS-FCOS (VOC) @128 MobileNetV2 · COCO test-devAP and FLOPs: 31.2 -> 32.0; 105.4 -> 39.3 GFLOPsImproved
replace FPN-FCOS by searched 128-256 decoder with improvementsNAS-FCOS ResNeXt-101 64x4d · COCO test-devAP: 44.7 -> 46.1Improved

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