AI Paper Atlas

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

TRACE: Your Diffusion Model is Secretly an Instance Edge Detector

Sanghyun Jo, Ziseok Lee, Wooyeol Lee, Jonghyun Choi, Jaesik Park, Kyungsu Kim

ICLR 2026 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Weakly supervised semantic segmentation · 2 citations (OpenAlex)

This paper tried a common change: Add a reconstruction / auxiliary loss

add reconstruction loss during distillation · COCO 2014 · AP^mk: 8.9 -> 9.4 Improved

What 987 accepted papers report for this change

See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.

Read the paper

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
apply ABDiv at a semantic timestep without IEP (ablation)TRACE on ProMerge · COCO 2014AP^mk: 3.1 -> 3.2No clear effect
add Instance Emergence Point timestep selection (ablation)ABDiv · COCO 2014AP^mk: 3.2 -> 4.8Improved
add one-step self-distillation with edge decoder (ablation)IEP + ABDiv · COCO 2014AP^mk: 4.8 -> 8.2Improved
add reconstruction loss during distillation (ablation)TRACE edge decoder · COCO 2014AP^mk: 8.9 -> 9.4Improved
use MSE instead of KL divergence for IEP (ablation)TRACE · COCO 2014AP^mk: 9.4 -> 3.8Hurt
use JSD instead of KL divergence for IEP (ablation)TRACE · COCO 2014AP^mk: 9.4 -> 9.4No clear effect
use non-diffusion backbones (ABDiv only) instead of diffusion backbones (ablation)TRACE · COCO 2014AP^mk: 7.1 (PixArt-alpha 0.6B) vs 4.1 (Qwen2.5-VL 72B)Hurt
use larger diffusion backbone (ablation)TRACE · COCO 2014AP^mk: 6.8 (SD1.5) -> 8.3 (FLUX.1)Improved
replace TRACE edges with conventional edge detectors in BGP (ablation)TRACE pipeline · COCO 2014AP^mk: 9.4 -> 1.2 (Canny), 4.3 (DiffusionEdge)Hurt
add TRACE edges to tag-supervised DHRDHR panoptic · VOC 2012 (ResNet-50)PQ: 45.0 -> 56.9Improved
use 8-neighborhood instead of 4-neighborhood in ABDiv (ablation)TRACE · COCOAP^mkNo clear effect

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