AI Paper Atlas

Research map / Detection and segmentation / CNN features

CODiff: One-Step Diffusion Model for Camouflaged Object Detection

Xiaotong Fu, Qian Liu, Qihang Zhou, Wenchao Meng, Qinmin Yang, Shibo He

ICML 2026 · Detection and segmentation › CNN features · approach: Context for hard detection conditions

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 global <cls> token and timestep embedding guidance (GSG) to the denoiser (ablation)CODiff · COD10Kweighted F-measure: 0.838 -> 0.852Improved
add timestep embedding alongside <cls> token (ablation)CODiff · NC4KMAE: 0.021 -> 0.020No clear effect
use multi-level DINOv2 features [3,6,9,12] instead of only final layer in HCI (ablation)CODiff · COD10Kweighted F-measure: 0.803 -> 0.852Improved
add a fourth DINOv2 layer (layer 3) beyond [6,9,12] (ablation)CODiff · NC4KS-alpha: 0.911 -> 0.911No clear effect
add representation alignment loss LRA with lambda 0.75 (ablation)CODiff · COD10KMAE: 0.017 -> 0.014Improved
raise LRA weight from 0.75 to 1.0 (ablation)CODiff · COD10KS-alpha: 0.896 -> 0.890Hurt
replace PVTv2 trainable encoder with frozen DINOv2 in a prior diffusion COD model (ablation)CamoDiffusion · COD10KS-alpha: 0.880 -> 0.870Hurt
use CODiff framework instead of CamoDiffusion with the same frozen DINOv2 backbone (ablation)frozen DINOv2 · COD10KS-alpha: 0.870 -> 0.896Improved
use one-step sampling instead of 10-step samplingdiffusion-based COD · COD10Ksampling time (s/img): 0.18 -> 0.07Improved
one-step CODiff versus the strongest transformer method ZoomNeXt-P on COD10K S-alphaCODiff · COD10KS-alpha: 0.898 -> 0.896Mixed

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