AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

CFCG: Semi-Supervised Semantic Segmentation via Cross-Fusion and Contour Guidance Supervision

Shuo Li, Yue He, Weiming Zhang, Wei Zhang, Xiao Tan, Junyu Han, Errui Ding, Jingdong Wang

ICCV 2023 · Label-efficient and robust learning › Semi-supervised learning · approach: Consistency training for semi-supervised segmentation · 25 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 coherent perturbation (image-level plus feature-level) to the baseline (ablation)Cross pseudo supervision baseline · PASCAL VOC 2012 1/8, ResNet-50mIoU: 73.08 -> 73.44Improved
add adaptive contour guidance module (ACGM) (ablation)CP baseline · PASCAL VOC 2012 1/8, ResNet-50mIoU: 73.44 -> 75.23Improved
add cross-fusion supervision (CFS) to CP baseline (ablation)CP baseline · PASCAL VOC 2012 1/8, ResNet-50mIoU: 73.44 -> 75.22Improved
combine ACGM and CFS together on top of CP (ablation)CFCG · PASCAL VOC 2012 1/8, ResNet-50mIoU: 73.08 -> 77.10 (single model)Improved
use fusion inference with both learners instead of a single learner (ablation)CFCG · PASCAL VOC 2012 1/8, ResNet-50mIoU: 77.10 -> 77.55Improved
train with CFS rather than average logits fusion of two learners (ablation)CFCG single learner · PASCAL VOC 2012 1/8, ResNet-50mIoU: 76.01 -> 77.10Improved
learn contour weight gamma with an exponential schedule vs constant 0 (no weight map) (ablation)ACGM · PASCAL VOC 2012 1/8, ResNet-50mIoU: 75.22 -> 77.10Improved
use log or constant-1 gamma schedule instead of exp (ablation)ACGM · PASCAL VOC 2012 1/8, ResNet-50mIoU: 77.10 -> 75.56 (log); 75.95 (constant 1)Hurt
use Gaussian kernel k=64 (img size/8) instead of 32 or 128 (ablation)ACGM · PASCAL VOC 2012 1/8, ResNet-50mIoU: 75.86 (k=32), 75.88 (k=128) -> 77.10Improved
emphasize unreliable contour regions (contour-strengthened) instead of down-weighting them (ablation)ACGM · PASCAL VOC 2012 1/8, ResNet-50mIoU: 75.01 -> 77.10Improved
use CFCG instead of CPS baselinePASCAL VOC 2012 ResNet-50, all partitions · PASCAL VOC 2012mIoU: +2.00% to +3.43% w/o fusion; +2.50% to +3.88% w/ fusionImproved

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