AI Paper Atlas

Research map / Detection and segmentation / CNN features

k-Means Mask Transformer

Qihang Yu, Huiyu Wang, Siyuan Qiao, Maxwell Collins, Yukun Zhu, Hartwig Adam, Alan Yuille, Liang-Chieh Chen

ECCV 2022 · Detection and segmentation › CNN features · approach: Plain and high-resolution ViTs · 123 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
replace spatial-wise softmax cross-attention with k-means cluster-wise argmax cross-attention (ablation)kMaX-DeepLab (ResNet-50) · COCO valPQ: 5.2% PQImproved
double object queries to 256 with drop query regularization (ablation)kMaX-DeepLab (ConvNeXt-L) · COCO valPQ: 57.9 -> 58.0Improved
double object queries to 256 without drop query regularization (ablation)kMaX-DeepLab (ConvNeXt-L) · COCO valPQHurt
add pseudo labels on COCO unlabeled set to training datakMaX-DeepLab (ConvNeXt-L) · COCO valPQ: 57.9 -> 58.1Improved
use Gumbel-Softmax instead of deep supervision to train through the non-differentiable argmax (ablation)kMaX-DeepLab · COCO valPQHurt
scale backbone from ConvNeXt-B to ConvNeXt-LkMaX-DeepLab · COCO valPQ: +0.7% for kMaX-DeepLab but +1.4% for Mask2FormerImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 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