AI Paper Atlas

Research map / Detection and segmentation / CNN features

MCIBI++: Soft Mining Contextual Information Beyond Image for Semantic Segmentation

Zhenchao Jin, Dongdong Yu, Zehuan Yuan, Lequan Yu

TPAMI 2022 · Detection and segmentation › CNN features · approach: Context modeling for semantic segmentation · 23 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 dataset-level context aggregation (MCIBI++) to FCN (ablation)FCN ResNet-50 · ADE20K valmIoU: 36.96 -> 43.39Improved
fuse by concatenation instead of element-wise add (ablation)FCN+MCIBI++ · ADE20K valmIoU: 41.15 -> 43.39Improved
update the memory module during training (ablation)FCN+MCIBI (Memory-R) · ADE20K valmIoU: 21.78 -> 42.18Improved
store distribution information (mean, variance) instead of raw representations (ablation)FCN+MCIBI++ · ADE20K valmIoU: 42.18 -> 43.39Improved
add coarse-to-fine iterative inference strategy (ablation)FCN+MCIBI++ · ADE20K valmIoU: 43.39 -> 43.74Improved
use fixed momentum 0.1 instead of polynomial annealing (ablation)FCN+MCIBI++ · ADE20K valmIoU: 42.45 -> 43.39Improved
set loss weight alpha to 0.4 instead of 1.0 (ablation)FCN+MCIBI++ · ADE20K valmIoU: 42.22 -> 43.39Improved
increase inference stages from 2 to 4 (ablation)FCN+MCIBI++ · ADE20K valmIoU: 43.74 -> 43.91 (49.33 ms -> 74.72 ms)No clear effect
initialize memory by cosine-similarity clustering instead of random pixel (ablation)FCN+MCIBI++ · ADE20K valmIoU: 43.39 -> 43.43No clear effect
add DCA to ASPP (DeepLabV3) as complementary context (ablation)DeepLabV3 ResNet-50 · ADE20K valmIoU: 43.19 -> 44.85Improved
add MCIBI++ to PSPNetPSPNet ResNet-50 · ADE20K valmIoU: 42.64 -> 43.89Improved
add MCIBI++ to UperNet trained on LIP, tested on CIHPUperNet · CIHP val (trained on LIP)mIoU: 26.84 -> 27.74Improved

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