AI Paper Atlas

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

Robust Trust Region for Weakly Supervised Segmentation

Dmitrii Marin, Yuri Boykov

ICCV 2021 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Weakly supervised semantic segmentation · 3 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 gradient descent on relaxed Potts loss with discrete trust region (alpha-expansion)Grid-GD · ScribbleSup val, full scribblesmIoU: 0.64 -> 0.67Improved
replace gradient descent on relaxed Potts loss with trust regionGrid-GD · ScribbleSup val, clicks (length 0)mIoU: 0.54 -> 0.57Improved
replace gradient descent on relaxed Potts loss with trust regionGrid-GD · ScribbleSup val, length 0.5mIoU: 0.62 -> 0.64Improved
use trust region on 8-grid Potts instead of dense CRF gradient descentDense-GD · ScribbleSup val, full scribblesmIoU: 0.64 -> 0.67Improved
add Potts regularized loss via trust region to partial cross-entropyPCE-GD · ScribbleSup val, full scribblesmIoU: 0.61 -> 0.67Improved
add Potts regularized loss via trust region to partial cross-entropyPCE-GD · ScribbleSup val, clicks (length 0)mIoU: 0.50 -> 0.57Improved
add Potts regularization with plain gradient descent (Grid-GD) to partial cross-entropyPCE-GD · ScribbleSup val, full scribblesmIoU: 0.61 -> 0.64Improved
use a fixed set of proposals (lambda=0) instead of lambda>0 trust region (ablation)Grid-TR · ScribbleSup, full scribblesmIoUHurt
use a very small trust region (lambda to infinity) instead of moderate lambda (ablation)Grid-TR · ScribbleSup, full scribblesmIoUHurt
use robust KL metric with outlier probability instead of plain cross-entropy (ablation)Grid-TR · ScribbleSupmIoUImproved
tune outlier probability epsilon to near the true noise levelrobust loss · Fashion-MNIST with noisy labels (half corrupted)accuracy: best at epsilon=0.4Improved

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