AI Paper Atlas

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

Few-Shot Semantic Segmentation With Cyclic Memory Network

Guo-Sen Xie, Huan Xiong, Jie Liu, Yazhou Yao, Ling Shao

ICCV 2021 · Detection and segmentation › Open-set and few-shot detection · approach: Few-shot segmentation · 73 citations (OpenAlex)

Read the paper

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 cyclic memory reading to multi-resolution baseline (ablation)Baseline · PASCAL-5i (4-fold mean)mean-IoU 1-shot: 59.9 -> 62.8Improved
add cyclic memory reading to multi-resolution baseline (ablation)Baseline · COCO-20i (4-fold mean)mean-IoU 1-shot: 36.0 -> 39.3Improved
add cyclic memory reading to multi-resolution baseline (ablation)Baseline · COCO-20i (4-fold mean)mean-IoU 5-shot: 39.7 -> 43.1Improved
add cyclic memory reading to multi-resolution baselineBaseline · PASCAL-5i (4-fold mean)FB-IoU 1-shot: 73.3 -> 72.3Hurt
add query feature re-adding (memory plus query value) (ablation)CMN w Memory Value · PASCAL-5i (folds 0-1, ablation)mean-IoU 5-shot: 66.7 -> 68.1Improved
add query feature re-adding (memory plus query value) (ablation)CMN w Query Value · PASCAL-5i (folds 0-1, ablation)mean-IoU 1-shot: 66.3 -> 67.2Improved
use ConvGRU recursive updating instead of concatenation (ablation)CMN w Concatenate · PASCAL-5i (folds 0-1, ablation)mean-IoU 1-shot: 65.8 -> 67.2Improved
use ConvGRU recursive updating instead of channel compression (ablation)CMN w Compress · PASCAL-5i (folds 0-1, ablation)mean-IoU 5-shot: 66.4 -> 68.1Improved
use three feature resolutions instead of two (ablation)CMN N=2 · PASCAL-5i (folds 0-1, ablation)mean-IoU 1-shot: 66.0 -> 67.2Improved
use four feature resolutions instead of three (ablation)CMN N=3 · PASCAL-5i (folds 0-1, ablation)mean-IoU 5-shot: 68.1 -> 67.6Hurt
use early feature fusion instead of mask-average late fusion (ablation)CMN 5-shot · PASCAL-5i (4-fold, 5-shot fusion)mean-IoU: 63.3 -> 63.7Improved
use early feature fusion instead of mask-OR late fusion (ablation)CMN 5-shot · PASCAL-5i (4-fold, 5-shot fusion)mean-IoU: 62.9 -> 63.7Improved

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