AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Unsupervised Object Segmentation by Redrawing

Mickaël Chen, Thierry Artières, Ludovic Denoyer

NeurIPS 2019 · Detection and segmentation › Open-set and few-shot detection · approach: Amodal and open-world instance segmentation · 61 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 information conservation loss (decode z from redrawn image) (ablation)ReDO · training stabilitynon-trivial segmentationImproved
redraw only a single region per iteration instead of all regions (ablation)ReDO · trainingmeaningful masksImproved
use Pyramid Pooling Module in mask network instead of standard residual network (ablation)ReDO mask network · segmentationmask qualityImproved
orthogonal initialization gain 0.8 vs 0.2 or 1.4 (ablation)ReDO · trainingsuccess of trainingMixed
use self-attention in D and G (ablation)ReDO · segmentationmask qualityNo clear effect
different ratio of discriminator to generator updates (ablation)ReDO · trainingmask qualityNo clear effect
unsupervised ReDO vs supervised model with same architecturesupervised F · LFWaccuracy/IoU: 92% accuracy and 79% IoU matched by supervised with 50-60 labelsMixed
train on fused Flowers+LFW without labels using Flowers hyperparametersReDO · Flowers+LFWIoU: accuracy 0.856, IoU 0.691No clear effect

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