AI Paper Atlas

Research map / Detection and segmentation / CNN features

Interaction via Bi-Directional Graph of Semantic Region Affinity for Scene Parsing

Henghui Ding, Hui Zhang, Jun Liu, Jiaxin Li, Zijian Feng, Xudong Jiang

ICCV 2021 · Detection and segmentation › CNN features · approach: Local features and context in CNNs · 10 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 graph-based region affinity reasoning (GRAr) (ablation)Dilated FCN ResNet-50 · PASCAL-Context ablationmIoU %: 42.3 -> 51.7Improved
add semantic region loss (SR-Loss) (ablation)FCN+GRAr ResNet-50 · PASCAL-Context ablationmIoU %: 51.7 -> 53.0Improved
add multi-scale testingGRAr+SR-Loss ResNet-101 · PASCAL-Context ablationmIoU %: 54.8 -> 55.7Improved
swap ResNet-101 for ResNest-101GRAr+SR-Loss+MS · PASCAL-Context ablationmIoU %: 55.7 -> 57.0Improved
use graph edges from co-occurrence statistics instead of learned Conv (positive only) (ablation)GRAr ResNet-50 · PASCAL-Context affinity ablationmIoU %: 48.1 -> 49.7Improved
use graph edges instead of learned Conv (positive and negative) (ablation)GRAr ResNet-50 · PASCAL-Context affinity ablationmIoU %: 48.2 -> 51.7Improved
add negative affinity to positive-only graph (ablation)GRAr ResNet-50 · PASCAL-Context affinity ablationmIoU %: 49.7 -> 51.7Improved
add positive affinity to negative-only graph (ablation)GRAr ResNet-50 · PASCAL-Context affinity ablationmIoU %: 49.3 -> 51.7Improved
GRAr module instead of OCR as plug-in on FCNFCN ResNet-50 · FCN+module comparison (PASCAL-Context)mIoU %: 50.4 -> 51.7Improved
GRAr module instead of PPM as plug-in on FCNFCN ResNet-50 · FCN+module comparison (PASCAL-Context)mIoU %: 49.2 -> 51.7Improved
GRAr vs OCR with ResNet-101GRAr · PASCAL-ContextmIoU %: 54.8 -> 55.7Improved
GRAr vs OCR HRNetV2-W48 with ResNest-101GRAr · COCO-Stuff (171 classes)mIoU %: 40.5 -> 42.6Improved

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