Title: Uncertainty-Edge Dual Guided Camouflage Object Detection
Paper link: Early access version avalible here.
[2023-07-11] Code and corresponding data is uploaded with user guide.
[2023-07-03] Paper has been accepted by IEEE Transaction on Multimedia. 🥳 Congradulations!!! 🥳
[2023-05-27] Detection results on four dataset: CHAMELEON, CAMO, COD10K-test, and NC4K are avilible: Drive or Baidu Netdisk(9cu6).
[2023-05-26] Initial repository.
[2022-11-22] Manuscript uploaded.
Figure 1: In this paper, we present the visualization results of the edge and uncertainty guidance operation in a highly challenging scenario. The red boxes indicate regions where the indistinguishable parts yield higher uncertainty scores. By incorporating edge information, our UEDG (Uncertainty Edge Guidance) approach achieves favorable performance.
-
Novel multi-task guided framework. We propose a novel structure that can combine multiple prior (uncertainty and edge in UEDG) for backbone feature guidance.
-
Powerful feature fusion strategy. We employed Uncertainty-Edge Mutual Fusion (UEMF), Uncertainty Deduce Module (UDM), Edge Estimate Module (EEM), and Uncertainty/Edge Guide Grouping (UGG/EGG) module with in a powerful end-to-end formation.
-
SOTA results. Our proposed method achieve the SOTA performance under four metrics in CHAMELEON, CAMO, COD10K, and NC4K. We also achieve the best performance in medical application like polyb segmentation as well.
Figure 2: UEDG strucuture overview.
Figure 3: UEDG strucuture overview.
conda create -n UEDG python=3.8
conda activate UEDG
pip install -r ./requirements.txt
For cuda11.6:
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge
- downloading testing dataset and move it into
./dataset/TestDataset/, which can be found in Drive. - downloading training dataset and move it into
./dataset/TrainDataset/, which can be found in Drive.
- preparing the pvt weights on PVT(Pyrimid Vision Transformer) refer to Drive and move it into
./pre_trained/.
python MyTrain.py
python MyTest.py
Code copied a lot from HVision-NKU/CamoFormer, thograce/BGNet, weijun88/F3Net, HUuxiaobin/HitNet, GewelsJI/DGNet, clelouch/BgNet, fanyang587/UGTR, whai362/PVT. Thanks for their great works!
The source code is free for research and education use only. Any commercial usage should get formal permission first.
@ARTICLE{10183371,
author={Lyu, Yixuan and Zhang, Hong and Li, Yan and Liu, Hanyang and Yang, Yifan and Yuan, Ding},
journal={IEEE Transactions on Multimedia},
title={UEDG:Uncertainty-Edge Dual Guided Camouflage Object Detection},
year={2023},
doi={10.1109/TMM.2023.3295095}}