AI Paper Atlas

Research map / Detection and segmentation / CNN features

Fast, Exact and Multi-scale Inference for Semantic Image Segmentation with Deep Gaussian CRFs

Siddhartha Chandra, Iasonas Kokkinos

ECCV 2016 · Detection and segmentation › CNN features · approach: Context modeling for semantic segmentation · 185 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
use 8-connected instead of 4-connected neighbourhood (ablation)QO · VOC12 valmIoU: 76.36 -> 76.40No clear effect
use 12-connected instead of 4-connected neighbourhood (ablation)QO · VOC12 valmIoU: 76.36 -> 76.42No clear effect
use shared Potts-type pairwise terms (ablation)QO · VOC12 valmIoU: 76.36 -> 76.59Improved
add one QO unit per image resolution (ablation)QO · VOC12 valmIoU: 76.36 -> 76.69Improved
couple scales with joint multi-resolution QO instead of decoupled per-resolution QO (ablation)QO^res · VOC12 valmIoU: 76.69 -> 76.93Improved
add QO module to basenetBasenet · VOC12 testmIoU: 72.72 -> 73.41Improved
add multi-resolution QO to basenetBasenet · VOC12 testmIoU: 72.72 -> 73.86Improved
add DenseCRF post-processing (ablation)QO^mres · VOC12 testmIoU: 73.86 -> 75.46Improved
add DenseCRF post-processing (ablation)Basenet · VOC12 testmIoU: 72.72 -> 73.78Improved
add CRF post-processing to Potts QO on Resnet-101 (ablation)QO^s Deeplab-v2 Resnet-101 · VOC12 testmIoU: 79.5 -> 80.2Improved
use QO^s instead of Deeplab-v2 with CRF, without CRFDeeplab-v2 Resnet-101 · VOC12 testmIoU: 79.7 -> 79.5No clear effect
replace parallel mean-field (Jacobi) with conjugate gradient solverQO inference · GPU inference speedspeed: up to 2.5 fasterImproved

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