AI Paper Atlas

Research map / Detection and segmentation / Weakly supervised detection and segmentation

Large-Scale Unsupervised Semantic Segmentation

Shanghua Gao, Zhong-Yu Li, Ming–Hsuan Yang, Ming‐Ming Cheng, Junwei Han, Philip H. S. Torr

TPAMI 2022 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Weakly supervised semantic segmentation · 107 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 non-contrastive P2P alignment (ablation)SwAV · ImageNet-S300 (distance matching)test mIoU: 22.6 -> 24.8Improved
add clustering P2P alignment (ablation)SwAV · ImageNet-S300test mIoU: 22.6 -> 21.2Hurt
add contrastive P2P alignment (ablation)SwAV · ImageNet-S300test mIoU: 22.6 -> 18.0Hurt
add D2S supervision at stage 3 (ablation)SwAV+P2P · ImageNet-S300test mIoU: 24.8 -> 25.2Improved
supervise stages 3-2 instead of stage 3 (ablation)SwAV+P2P · ImageNet-S300test mIoU: 25.2 -> 24.9Hurt
add D2S supervision at stages 3-2 (ablation)PixelPro+P2P · ImageNet-S300test mIoU: 22.0 -> 23.4Improved
add clustering loss (ablation)PixelPro · ImageNet-S300test mIoU: 15.8 -> 21.3Improved
use projection M in P2P (ablation)SwAV+P2P · ImageNet-S300Img-Acc: 57.1 -> 58.5Improved
use deep-to-shallow instead of same-stage supervision (ablation)PixelPro+P2P · ImageNet-S300test mIoU: 22.6 -> 23.4Improved
cross-view instead of same-view D2S supervision (ablation)PixelPro+P2P · ImageNet-S300test mIoU: 23.1 -> 23.4Improved
pixel-attention label generation instead of pixel-level (ablation)PASS_s · ImageNet-S50test mIoU: 12.7 -> 29.3Improved
add fine-tuning step (ablation)PASS_s · ImageNet-S50test mIoU: 26.0 -> 29.3Improved

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