AI Paper Atlas

Research map / Detection and segmentation / CNN features

Uncertainty-Guided Transformer Reasoning for Camouflaged Object Detection

Fan Yang, Qiang Zhai, Xin Li, Rui Huang, Ao Luo, Hong Cheng, Deng-Ping Fan

ICCV 2021 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 367 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 uncertainty-guided transformer (UGT) on top of UQN (ablation)UGTR · COD10K-TestS-measure: 0.730 -> 0.789Improved
add uncertainty-guided transformer (UGT) on top of UQN (ablation)UGTR · CAMO-TestS-measure: 0.728 -> 0.747Improved
add prototyping transformer (PT) to UGT (ablation)UGTR · COD10K-TestS-measure: 0.789 -> 0.818Improved
add prototyping transformer (PT) to UGT (ablation)UGTR · CAMO-TestS-measure: 0.747 -> 0.785Improved
replace UQN probabilistic network with deterministic ResNet50-FCN (ablation)UGTR · CAMO-TestS-measure: 0.742 vs 0.728No clear effect
use PT instead of global average pooling for high-level semantics (ablation)UGTR · COD10K-TestS-measure: 0.776 -> 0.811Improved
use PT instead of multi-scale pooling (ablation)UGTR · COD10K-TestS-measure: 0.792 -> 0.811Improved
add LPL loss to UGT + PT (ablation)UGTR · COD10K-TestS-measure: 0.811 -> 0.818Improved
add uncertainty-guided random masking (UGRM) vs no masking (ablation)UGTR · COD10K-TestS-measure: 0.797 -> 0.818Improved
use UGRM instead of standard random masking (ablation)UGTR · CAMO-TestS-measure: 0.775 -> 0.785Improved
increase number of prototypes T from 8 to 16 (ablation)UGTR · COD10K-TestS-measure: 0.806 -> 0.818Improved
increase number of prototypes T from 16 to 32 (ablation)UGTR · COD10K-TestS-measure: 0.818 -> 0.817No 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