[ECCV2024] Learning Camouflaged Object Detection from Noisy Pseudo Label (Poster)

This is the open-source repository for our paper Learning Camouflaged Object Detection from Noisy Pseudo Label, accepted at ECCV 2024!

Our Paper Can Be Seen at Paper

Framework Architecture

Performance

Training Process

Task Definition: Weakly Semi-Supervised Camouflaged Object Detection (WSSCOD)

We introduce a novel training protocol named Weakly Semi-Supervised Camouflaged Object Detection (WSSCOD), utilizing boxes as prompts to generate high-quality pseudo labels. WSSCOD leverages box annotations, complemented by a minimal amount of pixel-level annotations, to generate high-accuracy pseudo labels.

1. Dataset Division:

   • $\mathcal{D}_m = \{\mathcal{X}_m, \mathcal{F}_m, \mathcal{B}_m\}_{m=1}^M$: Pixel-level annotations $\mathcal{F}_m$, box annotations $\mathcal{B}_m$, and training images $\mathcal{X}_m$.
   • $\mathcal{D}_n = \{\mathcal{X}_n, \mathcal{B}_n\}_{n=1}^N$: Box annotations and images, where $M+N$ represents the number of training sets.

2. Training ANet:

   • Train ANet using dataset $\mathcal{D}_m$.
   • Use $\mathcal{B}_m$ as prompts and $\mathcal{F}_m$ for supervision.

3. Generating Pseudo Labels:

   • Use the trained ANet and dataset $\mathcal{D}_n$ to predict pseudo labels $\mathcal{W}_n$.

4. Constructing the Weakly Semi-Supervised Dataset:

   • Combine $\{\mathcal{X}_m, \mathcal{F}_m\}_{m=1}^M$ and $\{\mathcal{X}_n, \mathcal{W}_n\}_{n=1}^N$ to form $\mathcal{D}_t$.

5. Training PNet:

   • Train PNet using the dataset $\mathcal{D}_t$.
   • Evaluate performance with different $M$ and $N$ ratios:
     • PNet$_{F1}$: $M=1\%$, $N=99\%$
     • PNet$_{F5}$: $M=5\%$, $N=95\%$
     • PNet$_{F10}$: $M=10\%$, $N=90\%$
     • PNet$_{F20}$: $M=20\%$, $N=80\%$

Details: ANet and PNet Training

Aspect	ANet (Auxiliary Network)	PNet (Primary Network)
Stage	First	Second
Objective	Generate high-accuracy pseudo labels	Main camouflaged object detection
Data Input	Subset $\mathcal{D}_m$ with pixel and box annotations	Weakly semi-supervised dataset $\mathcal{D}_t$
Training Dataset	$\mathcal{D}_m = \{\mathcal{X}_m, \mathcal{F}_m, \mathcal{B}_m\}_{m=1}^M$	$\mathcal{D}_t = \{\mathcal{X}_m, \mathcal{F}_m\}_{m=1}^M \cup \{\mathcal{X}_n, \mathcal{W}_n\}_{n=1}^N$
Annotations	Pixel-level $\mathcal{F}_m$ and box $\mathcal{B}_m$	Pseudo labels $\mathcal{W}_n$ and pixel-level $\mathcal{F}_m$
Supervision	Pixel-level $\mathcal{F}_m$ for pseudo label generation	Pseudo labels $\mathcal{W}_n$ and pixel-level $\mathcal{F}_m$
Input Prompts	Box annotations $\mathcal{B}_m$ for camouflaged objects	Images $\mathcal{X}_m$ and $\mathcal{X}_n$
Performance Evaluation	-	Different settings: PNet$_{F1}$, PNet$_{F5}$, PNet$_{F10}$, PNet$_{F20}$
Training Goal	Generate high-quality pseudo labels $\mathcal{W}_n$	Improve detection accuracy with various $M$ and $N$ ratios

1. Download the Training and Test Sets

We have made the training and test sets available for download via the following links:

• Google Drive
• BaiDu Drive (Passwd: he37)

Once downloaded, place data.zip in the code/data directory and unzip it.

2. Train ANet

python code/TrainANet/TrainDDP.py --gpu_id 0 --ration 1 
# ration represents the proportion of pixel-level labels
# we find that one card training is better than four or eight cards

3. Generate Pseudo Labels

python code/TrainANet/Test.py --ration 1 
# ration represents the proportion of pixel-level labels

4. Train PNet

python code/TrainANet/TrainDDP.py --gpu_id 0 --ration 1 --q_epoch 20 --batchsize_fully 6 --batchsize_weakly 24 
# ration represents the proportion of pixel-level labels
# q_epoch means we change the q to 1 at this epoch 
# batchsize_fully means the number of fully annotated samples in a batch
# batchsize_weakly means the number of weakly annotated samples in a batch

5. Testing Process

python code/TrainPNet/Test.py --ration 1 
# ration represents the proportion of pixel-level labels

Pretrained Weights and COD Results

For ANet

We release the weight and prediction maps of $N=99\%$, $N=95\%$, $N=90\%$ and $N=20\%$ at Biadu Link.

For PNet

Model	Pretrained Weight	Prediction Description
PNet$_{F1}$	Google Link	$M=1\%$, $N=99\%$
PNet$_{F5}$	Google Link	$M=5\%$, $N=95\%$
PNet$_{F10}$	Google Link	$M=10\%$, $N=90\%$
PNet$_{F20}$	Google Link	$M=20\%$, $N=80\%$

References

@inproceedings{zhang2025learning,
  title={Learning Camouflaged Object Detection from Noisy Pseudo Label},
  author={Zhang, Jin and Zhang, Ruiheng and Shi, Yanjiao and Cao, Zhe and Liu, Nian and Khan, Fahad Shahbaz},
  booktitle={European Conference on Computer Vision},
  pages={158--174},
  year={2025},
  organization={Springer}
}